Site-specific spraying in chili pepper production requires organ-level segmentation of leaves, peppers, and flowers from handheld field images. However, RGB appearance becomes unreliable under organ overlap, occlusion, dust, and fruit specularity. Offline monocular depth from Depth Anything V2 provides an accessible structural prior without RGB-D sensing, but its reliability is spatially concentrated rather than uniform. To address this, we propose Depth-Routed Selective Attention (DRSA), an estimated-depth-guided segmentation network. DRSA predicts a single per-pixel routing field that, through one shared decision, jointly governs where depth-boundary cross-attention and residual depth fusion contribute, so geometric cues act near organ contours while RGB remains the default carrier. The routing field is calibrated online from depth-on and depth-suppressed predictions without trust-map annotation. We construct PepperField-EstDepth, a self-built dataset of 3, 940 handheld field images paired with estimated monocular depth, on which DRSA achieves \(90.20\%\) mIoU and \(84.48\%\) boundary mIoU, outperforming both RGB-only baselines and attention-based RGB-D fusion baselines; over the RGB segmentation reference, the gains are \(+1.98\) and \(+2.67\) percentage points, respectively. Under group cross-validation, DRSA reaches \(0.8919\pm 0.0031\) mIoU and \(0.8294\pm 0.0046\) boundary mIoU. For the spraying application, DRSA attains a target recall of 0.9814, a target precision of 0.9756, and an organ-level off-target activation of \(2.44\%\) . Visual-pipeline timing shows a segmentation-only latency of 43.0 ms under pre-generated estimated depth, rising to 219.6 ms for the RGB-to-mask visual pipeline when online Depth Anything V2-L depth generation is included. These results support DRSA as a pre-spray organ-level perception module.
Early and real-time detection of rice leaf diseases (RLD) poses a significant challenge for farmers, especially in rural regions with limited access to advanced technology. Conventional deep learning models often require substantial computational resources, rendering them impractical for deployment on mobile or edge devices commonly used in agricultural environments. Although models such as ResNet50, VGG16, and InceptionV3 can achieve high accuracy, they are computationally expensive and may be less suitable for real-time deployment in smart agricultural systems. Furthermore, many existing studies rely on limited datasets, which can restrict model generalizability under diverse real-world conditions. Although transfer learning can improve classification performance, developing lightweight models suitable for real-time IoT deployment remains challenging. To address these limitations, we curated a hybrid dataset of 7,092 rice leaf images by combining self-collected and Kaggle samples and proposed IoT-RiceMobileNet, a lightweight improved MobileNetV2-based model that can be effectively integrated into our developed IoT system. The proposed model outperformed transfer learning and deep learning baseline models, achieving 99.19% test accuracy and 99.20% precision while maintaining low computational complexity. We further confirmed model stability using stratified 5-fold and 10-fold cross-validation on both the constructed RLD and multi-source datasets, achieving mean accuracies of 98.04% and 98.13% on the constructed RLD dataset and 98.11% and 98.58% on the multi-source dataset, respectively. Moreover, our models compact size and 85.17 FPS inference speed support real-time deployment. Finally, the model was integrated into an IoT-enabled mobile, web, and cloud-based inference framework, demonstrating its practical potential for scalable rice disease detection in smart agriculture.
Reproduction assets foundThe paper's self-collected rice leaf image dataset and source code are publicly archived on Zenodo and GitHub, and the third-party Kaggle/Mendeley image datasets used to construct the RLD and multi-source datasets are publicly available. All are paper-specific, public, and actionable.Dataset · publicrice disease monitoring and decision support in precision agriculture.
Supporting information
S1 Appendix
Cross-validation results for the multisource dataset.
(PDF)
Data Availability
The self-collected rice leaf image dataset used in this study is publicly available through the archived IoT-RiceMobileNet repository on Zenodo: https://doi.org/10.5281/zenodo.21140529 . The public third-party datasets used in this study are available from their original sources: https://www.kaggle.com/datasets/dedeikhsandwisaputra/rice-leafs-disease-dataset/data
https://www.kaggle.com/datasets/anshulm257/rice-disease-dataset
https://data.mendeley.com/datasets/hx6f852hw4/2 Third-party images are not redistribOpen asset ↗Zenodo · 10.5281/zenodo.21140529lines:868-883Code · public://www.kaggle.com/datasets/anshulm257/rice-disease-dataset
https://data.mendeley.com/datasets/hx6f852hw4/2 Third-party images are not redistributed in the Zenodo archive and should be obtained from their original sources according to their respective licenses. The source code is publicly available through the GitHub repository: https://github.com/eimran/IoT-RiceMobileNet and is also archived on Zenodo at: https://doi.org/10.5281/zenodo.21140529 .
Funding Statement
The author(s) received no specific funding for this work.
References
1. Sokra I, Somaly S, Meta H, Sarun H, Molikoy C. Factors affecting rice production: A systematic review. J Agric Technol. 2026;2(1):19–46. doi: 10.6084/m9.figshaOpen asset ↗GitHub · eimran/IoT-RiceMobileNetlines:868-883Dataset · publicthe multisource dataset.
(PDF)
Data Availability
The self-collected rice leaf image dataset used in this study is publicly available through the archived IoT-RiceMobileNet repository on Zenodo: https://doi.org/10.5281/zenodo.21140529 . The public third-party datasets used in this study are available from their original sources: https://www.kaggle.com/datasets/dedeikhsandwisaputra/rice-leafs-disease-dataset/data
https://www.kaggle.com/datasets/anshulm257/rice-disease-dataset
https://data.mendeley.com/datasets/hx6f852hw4/2 Third-party images are not redistributed in the Zenodo archive and should be obtained from their original sources according to their respective licenses. The source code is Open asset ↗Kaggle · dedeikhsandwisaputra/rice-leafs-disease-datasetlines:868-883Dataset · publicataset used in this study is publicly available through the archived IoT-RiceMobileNet repository on Zenodo: https://doi.org/10.5281/zenodo.21140529 . The public third-party datasets used in this study are available from their original sources: https://www.kaggle.com/datasets/dedeikhsandwisaputra/rice-leafs-disease-dataset/data
https://www.kaggle.com/datasets/anshulm257/rice-disease-dataset
https://data.mendeley.com/datasets/hx6f852hw4/2 Third-party images are not redistributed in the Zenodo archive and should be obtained from their original sources according to their respective licenses. The source code is publicly available through the GitHub repository: https://github.com/eimran/IoT-RiceMOpen asset ↗Kaggle · anshulm257/rice-disease-datasetlines:868-883Dataset · publicived IoT-RiceMobileNet repository on Zenodo: https://doi.org/10.5281/zenodo.21140529 . The public third-party datasets used in this study are available from their original sources: https://www.kaggle.com/datasets/dedeikhsandwisaputra/rice-leafs-disease-dataset/data
https://www.kaggle.com/datasets/anshulm257/rice-disease-dataset
https://data.mendeley.com/datasets/hx6f852hw4/2 Third-party images are not redistributed in the Zenodo archive and should be obtained from their original sources according to their respective licenses. The source code is publicly available through the GitHub repository: https://github.com/eimran/IoT-RiceMobileNet and is also archived on Zenodo at: https://doi.org/10.5Open asset ↗hx6f852hw4/2lines:868-883Code / dataset availability confirmedOpenAlex · Crossref · checked 17 Sept 2026
Junjie Wu · Bei Zhou · Jie Liu · Zhuyang Xie · Jie Luo · Jiajun Liu · jiahui zhu · Mingju Li · Yan Ai · Mingwei Liu · Yu Chen
MaizeRapeseed / canolaSoybeanLiDAR / point cloudLeafStem / branch
Three-dimensional plant point clouds retain crop architecture, but organ-level phenotyping requires reliable semantic separation of leaves and stems. Plant-GeoAT is a geometry-aware parsing network that encodes local relative-XYZ neighbourhoods before RGB fusion and couples spatial neighbourhoods with feature–space relations for dense point prediction. We evaluated the model separately within the native protocol of a self-built structure-from-motion rapeseed dataset, an image-based soybean dataset, and laser-scanned Pheno4D maize and tomato datasets; these are within-dataset train/test experiments, not cross-dataset transfer or domain-generalisation tests. Across five seeds, mIoU was 92.26 ± 0.28%, 82.50 ± 0.24%, 99.74 ± 0.05%, and 94.75 ± 0.15%, respectively. Maize Stem IoU reached 99.57 ± 0.09%. Adding LLGE increased the four-dataset average mIoU from 66.38% to 85.84%, and the complete LLGE + SSCA model reached 92.31%. On the fixed six-sample test set, exploratory semantic-guided clustering achieved 86.67 ± 7.45% Count Accuracy; structural correctness ranged from 3/6 to 4/6 samples across seeds. Plant-height consistency was assessed independently on all 60 reconstructed rapeseed samples. The results indicate that geometry-to-context encoding produces organ-level semantic units for subsequent phenotyping, while independent instance-labelled datasets and additional growth stages are still needed for trait-level validation.
Reproduction assets foundThe paper's Plant-GeoAT research-preview code (model implementation, training and evaluation scripts) is publicly available at an immutable GitHub commit under the authors' kimevans111 account. The self-built rapeseed point-cloud dataset and annotations are not public (available only on request), and third-party publicCode · publictheir original providers and publications; this article does not redistribute
third-party datasets. The processed summary metrics required to evaluate the reported results are
included in the main article and Supplementary Material. The reviewed Plant-GeoAT research-
preview code is available at the immutable historical commit https://github.com/kimevans111
/Plant-GeoAT/commit/e0aba5b94e26ba4306d0f7f9901185883fe8d4c4 (accessed on 30 August 2026).
This snapshot is distinct from the repository’s current default branch. It contains the model implemen-
tation, training and evaluation scripts, data-format and release documentation, command templates,
citation metadata, and licence/notiOpen asset ↗Plant-GeoATpdf-raw-page:24 lines:1-50Plant phenotyping relevance matchEurope PMC · checked 17 Sept 2026
Cell division patterns shape developing organs, but their systematic analysis remains limited by the technical difficulty of existing detection methods. Here, we present a practical method for inferring recent daughter-cell pairs from static cell shapes. Although cell divisions have long been inferred by expert judgment, the accuracy of this approach has rarely been quantified. Using live imaging of Arabidopsis leaf primordia, we evaluated 49 features from 7,573 neighboring cell pairs, including 1,114 direct daughter pairs and 6,459 non-daughter pairs. A simple junction-angle criterion correctly inferred 1,030 daughter pairs with 94.1% precision and 92.5% recall, whereas symmetry-based indices performed poorly. Temporal analysis revealed that junction angles were initially high after division, decreased during growth, and were altered by subsequent divisions, thus, separating daughter pairs from non-daughter pairs. This method also achieved 98.4% precision and 97.2% recall in an independent shoot apical meristem live-imaging dataset. Application to the pointed-tip Arabidopsis mutant rpl4d-3 and two non-model species further supported its potential applicability for analyzing division patterns in diverse leaf morphologies. For community use, we provide FIJI/ImageJ and Qt-based graphical user interface implementations. Overall, this study quantitatively validates empirical geometric cues and provides a practical framework for inferring daughter pairs from static tissue images.
Premise Herbarium specimens are increasingly used to extract morphological traits for ecological and evolutionary studies, yet the effects of tissue desiccation on trait measurements remain poorly understood. Here, we tested two hypotheses: (H1) whether higher tissue water content leads to greater measurement changes after herborization and (H2) whether fresh trait values can be reliably predicted from measurements of herborized specimens. Methods We evaluated the reliability of herbarium-based measurements by comparing fresh and dried traits of leaves, flowers, fleshy fruits, and seeds across 262 individuals representing 133 Neotropical Myrtaceae species. Phylogenetic least squares models and machine learning regressions were used to test H1 and H2. Results Leaves and flowers generally shrank after herborization, fruits size metrics tended to increase, and seeds were largely unaffected. Water content was significantly associated with the magnitude of herborization effects in flowers and some leaf and seed traits. Fresh trait values were accurately predicted from measurements of herborized specimens. Prediction errors were lowest for leaf traits, followed by fruits, flowers, and seeds. Discussion These results partially support H1 and support H2, indicating that herborized specimens can be reliably used for trait analyses when organ-specific responses are considered, providing a practical framework to account for potential desiccation bias in functional trait research.
Reproduction assets foundThe paper's trait measurements (Appendix S6 species mean data) and analysis code are openly available in the authors' GitHub repository, explicitly stated in the Data Availability Statement.Code · publice also acknowledge the Sociedade Botânica do Brasil (SBB) for financial support through the Scientia Amabilis 2024 grant.
DATA AVAILABILITY STATEMENT
The data supporting the findings of this study are available in the Supporting Information of this article (Appendix S6 ). Additionally, all data and code are openly accessible at https://github.com/ykilsztajn/fresh_dry_myrtaceae .
REFERENCES
Bewley , J. D.
,
K. J.
Bradford
,
H. W.
Hilhorst
, and
H.
Nonogaki
. 2012 .
Development and maturation
. In
Seeds: Physiology of development, germination and dormancy , 3rd ed. , 27 – 83 .
Springer
,
New York, New York, USA
.
Blonder , B.
,
V.
Buzzard
,
I.
Simova
,
L.
Sloat
,
B.
Boyle
,
R.
Lipson
,
B.
AguiOpen asset ↗ykilsztajn/fresh_dry_myrtaceaelines:268-513Plant phenotyping relevance matchCrossref · checked 17 Sept 2026
Published15 Sept 2026Black Sea Journal of Agriculture
Leaf area (LA) is one of the most important morphological traits for the assessment of plant growth, biomass production and physiological performance. For forage crops, rapid, accurate and non-destructive methods of LA estimation are particularly important to allow repeated measurements throughout the growing season. The aim of this study was to compare the leaf area estimation of sainfoin (Onobrychis viciifolia Scop.) by four regression models (linear, logarithmic, polynomial and multiple linear regression) based on simple plant morphological measurements. Ninety sainfoin plants were evaluated at the flowering stage under controlled conditions. Plant height and canopy width were measured manually and actual leaf area was measured with a LI-COR LI-3100C Leaf Area Meter. Regression analyses were performed using SPSS software to construct prediction models and compare the performance of the models. Of the models tested, the multiple linear regression model had the highest prediction accuracy (R²=0.981), which was significantly better than the polynomial (R²=0.413), linear (R²=0.399) and logarithmic (R²=0.361) models. The results demonstrated that the combination of plant height and canopy width greatly improved the leaf area estimation over the use of a single predictor. The predictive performance of simple regression models was low. Multiple linear regression was a reliable, rapid and non-destructive method to estimate the leaf area of sainfoin. These results pave the way for practical morphological models that could assist plant growth monitoring and agronomic studies while limiting the number of destructive samplings.
Nadeem Fareed · Carlos Alberto Silva · Alexander J. Gaskins · Susan J. Prichard · Andrew T. Hudak · Jinyi Xia · Cesar Ivan Alvites Alvites Diaz
Field / plotLiDAR / point cloudLeafStem / branchMorphology / geometry measurementSegmentationArchitecture / morphology / geometry
Terrestrial Light Detection and Ranging (LiDAR) provides detailed three-dimensional (3D) observations of forest structure, yet transferable wood–foliar semantic segmentation remains challenging because of forest structural heterogeneity, occlusion, and variability across forest ecosystems and terrestrial LiDAR platforms. Existing Deep Learning (DL) approaches are commonly developed for localized forest conditions or complex multi-class semantic taxonomies, often limiting their transferability to structurally distinct forests. Here, we introduce points2SBL, a geometry-driven, terrestrial LiDAR platform-agnostic DL framework that reduces heterogeneous forest vegetation to a transferable wood–foliar representation and subsequently uses these semantics as the foundation for 3D forest fuel characterization. The framework was developed using globally harmonized benchmark datasets acquired across multiple terrestrial LiDAR systems and broadleaf, coniferous, mixed, regenerating, and structurally complex forests. Three point-based architectures (PointNet++, PointNeXt, and Point Transformer) were benchmarked under identical training and evaluation protocols. Point Transformer provided the most consistent cross-dataset performance, achieving overall accuracy (OA) of 92–94%, mean Intersection-over-Union (mIoU) of 0.78–0.87, macro F1-score of 0.87–0.93, and Matthews correlation coefficient (MCC) of 0.74–0.86 across five independent benchmark datasets. Species-level, vertical-profile, and qualitative cross-ecosystem assessments further demonstrated consistent preservation of wood–foliar semantics throughout the 3D forest fuel continuum, including previously unseen forest point clouds, while revealing that some apparent semantic disagreements originated from manual annotation of benchmarks. Finally, the points2SBL wood–foliar semantics were geometrically decomposed using a geometric algorithm named “Points2woodyseg” into fuel-related components: stem, branches, leaf, surface wood, and surface foliage. Collectively, these results demonstrate that a simplified wood–foliar abstraction can provide a transferable semantic foundation for characterizing structurally heterogeneous terrestrial LiDAR point clouds and deriving essential 3D forest fuel information across diverse forest ecosystems.
Introduction Accurate extraction of maize seedling phenotypes is essential for early growth assessment, yet field-grown seedlings exhibit complex backgrounds, non-rigid leaf deformation, self-occlusion, and adhered leaves that challenge three-dimensional point-cloud analysis. Methods Here, we developed a workflow integrating PMSPA-Net, an enhanced PointNet++ model, with geometric region decomposition for semantic and organ-level segmentation of field-grown maize seedlings. Multi-view RGB images of 65 complete plants were reconstructed with OpenMVS, and all data were partitioned at the whole-plant level before point-block sampling. Fifteen plants were retained as a fixed independent test set, while the remaining samples were used for training and validation across three predefined random seeds. Results Under a common protocol, PMSPA-Net outperformed PointNet, PointNet++, DGCNN, and RandLA-Net, achieving an mIoU of 0.8286 ± 0.0149, a macro-F1 of 0.8981 ± 0.0238, and a balanced accuracy of 0.9561 ± 0.0223. A four-configuration ablation study showed complementary gains from the multi-scale spatial pyramid attention module and IOA-based hyperparameter optimization. Geometric region decomposition also achieved more accurate and complete adhered-leaf separation than DBSCAN in three annotated cases. Point-cloud-derived leaf length and width agreed well with manual measurements. Discussion This workflow demonstrates the feasibility of integrated semantic and organ-level phenotyping under the evaluated field conditions; validation across additional sites, developmental stages, and acquisition conditions is required before broader generalization.
Abstract Leafhopper feeding-damage severity is an important factor in the raw-material grading and final quality of Zijin Chan tea, but conventional assessment relies heavily on subjective visual judgment. This study developed a two-stage workflow combining machine-vision screening with offline near-infrared (NIR) spectroscopic reassessment. First, an improved small You Only Look Once version 8 (YOLOv8s) model localized tea-leaf targets and classified slight, moderate, and severe feeding damage. Physical samples corresponding to machine-vision no-result targets were then reassessed using an NIR model integrating standard normal variate preprocessing, Pearson correlation-based spectral-region selection, the successive projections algorithm, and a support vector machine. On 200 test images containing 4,000 annotated targets, the improved YOLOv8s model achieved precision of 88.6%, recall of 87.7%, and mean average precision at an intersection-over-union threshold of 0.5 of 89.3%, representing improvements of 12.2, 9.3, and 8.7 percentage points over the baseline, respectively. Developed from 240 independent physical samples, the NIR model achieved macro-averaged recall of 94.03% on the cross-validated training set and 95.12% on the spectral model-selection set. When the fixed model was applied to physical samples corresponding to 492 machine-vision no-result targets, 473 targets were classified correctly, yielding a reassessment accuracy of 96.14%. Across all targets, standalone machine vision achieved an end-to-end accuracy of 77.7%, whereas the two-stage workflow achieved 89.5%. These results support the feasibility of combining visual and spectral information for laboratory-based grading. Validation using in situ spectra and samples from multiple batches and seasons remains necessary before online deployment.
Diseases that affect plants considerably lower the production of crops and pose a significant threat to global food security, particularly in high-value crops like bell pepper (Capsicum annuum L.). The timely and sensitive identification of the disease is essential in reducing losses related to yield and paving the way to sustainable management of crops. Although automated plant disease diagnosis has been improved by artificial intelligence (AI)-driven image analysis, single-model methods are not always robust or generalisable. The study proposes a hybrid convolutional neural network-random forest (CNN-RF) framework for the early detection of bell pepper leaf diseases. Deep visual features of leaf images were extracted and subsequently used with a pretrained ResNet50 model and then used to classify the images with the help of a RF ensemble algorithm. Experiments were conducted on a publicly available PlantVillage bell pepper dataset comprising 495 leaf images (199 bacterial spot and 296 healthy samples). The classification accuracy of the proposed hybrid model was 97.98 % with high class-wise precision, recall and F1-scores. Receiver operating characteristic (ROC) analysis produced an area under the curve (AUC) of 0.98, indicating excellent discriminative performance. The dataset was evaluated using a stratified 80 : 20 train-test split. The results show that deep feature extraction combined with an ensemble learning method can improve classification accuracy and offer a scalable solution to assist in precisionagriculture and timely disease control. However, the framework was evaluated only on the PlantVillage dataset and has not yet been validated under real-field agricultural conditions. The framework demonstrates potential for extension to multi-disease classification tasks, although further validation on field-acquired datasets is required.
Leaves of terrestrial plants possess heterogeneous anatomical structures composed of multiple cell layers. Although biochemical differences among leaf tissues have been inferred from protein analyses and anatomical studies, direct comparisons of dynamic photochemical responses among tissues while preserving their spatial context remain challenging. Leaves experience intrinsically heterogeneous environments because incident light enters primarily from above and propagates through complex internal leaf structures. Therefore, analyzing photosynthetic activity at the tissue and cellular levels is essential for understanding how photosynthesis operates within structurally heterogeneous leaves. Here, we combined live leaf-section imaging with microscopic Imaging-PAM chlorophyll fluorescence measurements to analyze photochemical responses at the tissue level in living leaf sections. In dorsiventral dicot leaves, palisade tissues exhibited a higher effective PSII quantum yield [Y(II)] and more rapid induction of regulated energy dissipation [Y(NPQ)] than spongy tissues, indicating higher photosynthetic capacity and photoprotective activity. In contrast, rice leaves, which lack palisade-spongy differentiation, showed uniform photochemical responses along the adaxial-abaxial axis. In the C4 plant finger millet, mesophyll and bundle sheath cells displayed distinct photochemical responses consistent with their functional differentiation in C4 photosynthesis. These results demonstrate that photosynthetic responses are spatially organized according to leaf anatomical structure. Although section-based measurements do not reproduce the native optical and gas environments of intact leaves, they enable the comparison of intrinsic tissue-specific photochemical properties under approximately equivalent illumination. The present approach enables tissue-resolved chlorophyll fluorescence analysis in living leaves, providing a new framework for investigating how leaf architecture shapes the spatial organization of photosynthetic activity.
This study presents a systematic quantitative multi-elemental investigation of four major plant organs (roots, stems, leaves, and flowers) of Nerium oleander using calibration-based Laser-Induced Breakdown Spectroscopy (LIBS), with the results validated by Atomic Absorption Spectroscopy (AAS). Plasma characterization was carried out using Boltzmann plot and Stark broadening analyses, while negligible self-absorption observed through the Hα emission line confirmed optically thin plasma conditions and reliable quantitative measurements. A total of nine elements were detected, including Fe, Zn, Mn, Ca, Mg, K, Na, Cu, and Ni, each exhibiting different concentration levels across the analyzed tissues. Compositional analysis using standard calibration curve-based LIBS demonstrated that elemental concentrations were non-uniform, showing marked variations between the different plant tissues. Among the detected elements, calcium emerged as the most prevalent across all tissues. The highest calcium concentration was observed in leaves (16,385 mg L-1), followed by roots (12,092 mg L-1) and flowers (11,185 mg L-1). Root tissues exhibited elevated concentrations of Fe and Mn, reaching 1918 and 513 mg L-1, respectively. In contrast, flowers showed the highest Mn concentration (1888 mg L-1), while leaves were enriched in Mg (4445 mg L-1) and K (3700 mg L-1). The highest Na concentration was observed in stems (8025 mg L-1). Trace metals, including Cu, Zn, and Ni, were detected at comparatively low concentrations, while Pb remained undetected in all samples. The strong agreement between LIBS and AAS measurements confirms the reliability of the proposed methodology and demonstrates the potential of LIBS as a rapid and non-destructive tool for elemental assessment of medicinal plants.
Soil contamination represents a growing environmental concern, particularly in areas such as the "Land of Fires" (Campania region, Italy). This study evaluates the potential of Zea mays L. as a bioindicator of soil contamination through a multiscale approach. In bin-like containers under natural conditions, maize plants were grown on untreated (C) and artificially contaminated (T) soil with a mixture of heavy metals (Pb, Zn, Cr) and benzo[a]pyrene. Analyses were conducted at leaf, canopy, and field scales, integrating eco-physiological, spectral, and UAV measurements. At the leaf level, contamination induced functional alterations in photochemical and spectral traits that were strongly dependent on the growth stage, being most pronounced early in the season and progressively attenuating thereafter. At the canopy level, contaminated plants developed less height and leaf area, together with lower chlorophyll- and canopy-density-sensitive spectral indices, consistent with reduced vegetative vigor under contamination stress. Solar Induced Fluorescence (SIF) showed no residual difference between treatments once normalized for biomass production at the end of the growing season, in agreement with leaf-level active fluorescence measurements taken on the same date. At the field scale, UAV-based classification achieved superior performance with hyperspectral relative to multispectral data in distinguishing contaminated areas. Taken together, these results show that, under the experimental exposure conditions adopted here, the detectability of soil pollution by maize progressively shifts from early, transient physiological signals to persistent structural differences, while UAV hyperspectral data provided the highest discrimination between treatments. By tracking how leaf-level functional alterations translate into canopy- and field-scale structural separability, this study provides a scale-explicit assessment of maize as a promising bioindicator of soil contamination.
The precise temporal characterization of early growth dynamics in crops under abiotic stress is critical for stress resistance management in smart agriculture. However, traditional manual measurement methods struggle to achieve high-throughput and accurate quantification of multiple phenotypic parameters over continuous time series. To address this, we developed a full-time-series crop growth monitoring system that continuously collects fine-grained image data of cucumber seed germination and seedling growth under soil culture conditions, constructing an annotated dataset for identifying germination status and quantifying cotyledon area. The YOLOv8n-F_SGX and YOLOv8-seg models were developed and deployed, achieving detection and segmentation accuracies of 0.978 and 0.987, respectively, which enabled automatic monitoring of germination rate and precise extraction of cotyledon area. To further investigate the mitigating effects of nanomaterial priming on salt stress, we conducted germination and seedling growth experiments on cucumber seeds using a series of ZnO-NPs (particle size 30 nm) suspension concentrations at different salt stress levels. The results demonstrated that under salt stress conditions of 0–150 mmol·L −1 , priming with 100 mg·L −1 ZnO-NPs resulted in optimal germination dynamics (highest germination rate and fastest germination speed), along with the largest cotyledon area and highest growth rate in seedlings. In contrast, when the ZnO-NPs concentration increased to 400 mg·L −1 or higher, it significantly inhibited seed germination and seedling growth.
Lois Onyejere Nwobodo · Chinonso J. Okonkwo · Edith Angela Ugwu · Udeh Chukwuma Callistus · Okorie Kingsley Maduabuchi · Ngene John Ndubisi · Agbo Kenechukwu Martin
Field / plotLeafClassificationStress / disease detectionDisease symptoms / severity
Plant diseases threaten agricultural productivity, and automated image analysis can support early identification of visible disease symptoms. This study introduces HybOptic-CNN, a convolutional neural network (CNN) whose learning rate and batch size are selected using a hybrid Whale Optimization Algorithm--Grey Wolf Optimizer (WOA-GWO). Nine disease classes were selected from the 22-class CCMT field-image dataset, and 152 local farm leaf images were collected in Enugu State, Nigeria. Of the local images, 122 (80.3%) were added to the model-development data for training and validation, whereas 30 (19.7%) formed an independent Nigerian hold-out set excluded from augmentation, class balancing, early stopping, validation, and hyperparameter selection. Across 10 model-development runs, the optimized model achieved 96.8 ± 0.4% mean validation accuracy, 95.2 ± 0.5% macro-precision, 94.9 ± 0.6% macro-recall, and 95.0 ± 0.5% macro-F1, compared with 90.3 ± 0.9% validation accuracy and 86.3 ± 1.2% macro-F1 for the baseline. The optimized model improved mean validation accuracy by 6.5 percentage points and converged 14.6 epochs earlier. On the independent 30-image Nigerian hold-out, HybOptic-CNN achieved 93.3% accuracy and 93.1% macro-F1 across four represented disease classes. A web application integrating the trained classifier was also demonstrated. These results support improved model-development performance through hybrid hyperparameter selection and motivate broader multi-location field evaluation.
Reproduction assets foundThe paper's Data availability statement points to two public sources: a Mendeley dataset (the locally collected Nigerian field images) and the Kaggle CCMT plant disease dataset used as the principal image source. Only the Kaggle URL matches an allowed URL; the Mendeley URL is not in the allowed list, so only the CCMT/KDataset · publicnt and independent field-test data and should pub-
lish the class-wise split manifest, random seeds, WOA-GWO
numerical settings, and evaluation code so that the reported pro-
cedure can be reproduced and extended.
Data availability
The data used in this study are available at https://
data.mendeley.com/datasets/bwh3zbpkpv/1 and https://www.kaggle.com/datasets/rahimanshu/ccmt-plant-disease-dataset.Declaration of competing interest
The authors declare that they have no known competing fi-
nancial interests or personal relationships that could have ap-
peared to influence the work reported in this manuscript.
Funding
The authors received no specific funding from any public,
commercial, or not-fOpen asset ↗Kaggle · rahimanshu/ccmt-plant-disease-datasetpdf-raw-page:12 lines:1-78Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Published11 Sept 2026Journal of visualized experiments : JoVE
Volatile organic compounds (VOCs) released by plant leaves play key roles in stress signaling, plant-atmosphere interactions, and plant defense. Among these, wound-induced VOCs (wVOCs) are emitted within seconds of mechanical damage, herbivory, or environmental disturbance. Their emission dynamics depend strongly on the timing, severity, and method of tissue disruption. Yet, accurate quantification remains challenging due to mechanical artifacts, variable exposure conditions, and delays between injury and measurement. This study presents a standardized within-chamber leaf excision protocol for real-time monitoring of wVOCs and gas exchange. A surgical-grade cutter was integrated into a portable gas-exchange chamber to enable clean, controlled cuts within a sealed chamber under stable light, humidity, CO2, and temperature conditions. A proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF-MS) continuously measured volatile emissions at the chamber outlet, minimizing delay and signal distortion. This setup resolves emission onset, peak timing, maximum rise rate, and total release with high temporal fidelity. Application of the method to Quercus rubra, Acer platanoides, and Gossypium hirsutum demonstrated its ability to resolve distinct wound-induced emission patterns across contrasting leaf types. By eliminating delays associated with conventional sampling, this method resolves the full kinetic trajectory of wound-induced emissions and overcomes major limitations of previous approaches. It provides a robust framework for studying rapid stress responses in plant physiology, ecological biochemistry, and plant-atmosphere interactions.
Transmission electron microscopy (TEM) is widely used to examine plant cellular ultrastructure, but sample preparation remains challenging because polysaccharide-rich cell walls, large vacuoles and complex membrane systems compromise staining and structural preservation. Here, we developed a plant-adapted TEM preparation method based on a modified osmium-thiocarbohydrazide-osmium (OTO) staining workflow combined with optimized washing, dehydration and resin infiltration. Using tomato roots and leaves, we show that the optimized method produces cleaner backgrounds, better-defined cellular boundaries and improved preservation of cellular morphology compared with conventional preparation. Fine membrane-associated structures, including mitochondrial cristae, chloroplast grana and stroma lamellae, nuclear membranes and other endomembrane structures, were more clearly resolved, while staining-related precipitates and diffuse background artifacts were reduced. This method provides a practical approach for high-contrast TEM imaging of plant tissues and ultrastructural analysis of plant cells and organelles.
Reliable prediction and interpretation of photosynthetic rate are important for precision environmental management in greenhouse tomato production, but model stability is often limited by variable redundancy, measurement-session effects, and environmental heterogeneity. This study developed an integrated gas-exchange-data-based framework for key-factor selection, photosynthetic-rate reconstruction, cross-session validation, environmental correction, and physiology-informed limitation screening. Core predictors were first identified from high-dimensional gas-exchange variables using K-means clustering and random-forest importance analysis. A CatBoost–ExtraTrees–RBF-SVR stacked ensemble was then constructed to reconstruct the leaf photosynthetic rate from selected gas-exchange variables, and its cross-session generalization was evaluated using nested cross-validation and leave-one-file-out (LOFO) extrapolation. Environmental correction was further applied to improve cross-session comparability, and rule-based limitation screening was used to classify potential environmental constraints associated with reduced photosynthesis. The stacked model achieved an RMSE of 0.806 and an R2 of 0.9918 under nested cross-validation, and an RMSE of 0.814 and an R2 of 0.9917 under LOFO extrapolation. After environmental correction, the cross-session variability of the environmentally corrected photosynthetic indicator was reduced by 96.0%, reflecting improved cross-session comparability. Under deployable sensor inputs, performance decreased markedly (A1: RMSE = 4.578, R2 = 0.735; A2: RMSE = 4.610, R2 = 0.731), compared with the gas-exchange baseline (RMSE = 0.833, R2 = 0.991). Thus, routine greenhouse sensor models should be regarded as approximate rather than high-accuracy substitutes for the gas-exchange-based model.
Reproduction assets foundThe paper analyzes a publicly available greenhouse tomato gas-exchange dataset (Manjarrez-Sanchez & Martinez-Carrillo, Data Brief 2020), which is the exact phenotype/sensor input data for this study's photosynthesis reconstruction analysis. The authors' own processed data and analysis scripts are only available upon请求,Dataset · publicThe data used in this study were obtained from the publicly available greenhouse
tomato gas-exchange dataset reported by Manjarrez-Sanchez and Martinez-Carrillo [29].Open asset ↗pdf-raw-page:4 lines:1-38Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Zea mays (maize) is a critical global crop, generating more than 1 billion metric tons of grain annually. Plant morphology is tightly associated with final yield and is determined by the process of development. Accurate assessment and recording of mature plant morphology is essential for linking genotype to phenotype and for assessing environmental influences on plant development and yield. In maize, imaging of fully emerged aerial organs provides an accessible and informative developmental end point for comparative phenotypic analyses. Here, we provide a protocol for capturing high-quality images of mature maize plants and detached leaf blades, suitable for publication and downstream quantitative trait extraction. We emphasize standardized imaging conditions, including high-contrast backgrounds, diffuse lighting, fixed camera positioning, and consistent scale calibration, to minimize technical variation and to facilitate automated image processing. Guidance is provided for both laboratory- and field-based imaging setups and uses widely available equipment. Images generated using this protocol can be analyzed using platforms such as FIJI to extract morphometric traits, including length, width, and outline-based shape descriptors. By prioritizing standardization at the point of acquisition, this protocol supports reproducible, scalable phenotyping of mature maize morphology for developmental studies in maize.
Background and aims Recognizing lineages is a central challenge in plant systematics, making it essential to explore multiple analytical tools. In this context, this study investigates how frond shape can assist in discriminating against lineages within the Scaly clade of Microgramma (Polypodiaceae), and tests whether the integration of multiple lines of evidence enables a more consistent recognition of lineages than exclusively macromorphological approaches. Methods We analyzed 271 specimens representing eight species, using Elliptical Fourier Analysis (EFA) to quantify frond shape, followed by multivariate statistical tests (PCA, MANOVA, LDA). Evolutionary relationships between spectral and morphometric data were assessed through phylogenetic generalized least squares (PGLS) regressions and phylogenetic partial least squares (Phylo-PLS) analyses. Results Dimorphic species exhibited higher discrimination capacity (average accuracy of 80-83%). Fertile and combined fronds yielded the highest accuracy values. Morphologically similar species, such as M. reptans and M. tobagensis, showed considerable overlap, whereas M. percussa achieved the best performance (average accuracy of 80%). Morphometric-spectral covariation yielded R2 = 0.72 (P = 0.003) when sterile and fertile fronds were combined, although PGLS analyses detected no significant phylogenetic signal. The association differed between frond types and is interpreted here as exploratory evidence of morpho-spectral covariation. Conclusions Outline morphometry combined with infrared spectroscopy within a phylogenetic framework improves lineage discrimination, although overlap zones persist, reflecting complex evolutionary processes. Our study highlights the potential of integrative systematics to elucidate species boundaries in groups with high morphological disparity, as well as the need for broad sampling and multi-evidence approaches in future systematic reviews.
Abstract Image-based plant-disease classifiers often achieve high accuracy under controlled conditions, but their reliability can deteriorate in real-world field environments. This article investigates crossdataset strawberry leaf-health classification using controlled PlantVillage images and field images collected through the SmartBerry project in Nigeria. Four deep-learning architectures were evaluated under within-domain, controlled-to-field and pooled-training conditions using healthy-leaf and darkspot/ scorch symptoms. Although all models achieved near-ceiling performance on PlantVillage, direct transfer to SmartBerry reduced macro F1-score to between 0.418 and 0.824, demonstrating a substantial gap between controlled-dataset performance and field reliability. To reduce the amount of locally labelled field data required to address this gap, this paper proposes SmartBerry Domain Adaptation (SmartBerry-DA). This label-efficient adaptation approach combines a model trained on controlled images with a small set of labelled SmartBerry images and additional unlabelled field images. Using MobileViTv2-0.5, SmartBerry-DA achieved a macro F1-score of 0.940 ± 0.015 with only five labelled SmartBerry images per class and 0.988 ± 0.012 with ten, compared with 0.994±0.008 when the full labelled target dataset was used. This shows that strong field performance can be recovered with substantially reduced local annotation. For lightweight offline deployment, 16- bit floating-point representation reduced model size from 4.37 MB to 2.30 MB without changing predictive performance under the evaluated conditions. The findings demonstrate the importance of locally acquired field data and label-efficient adaptation for transferring agricultural AI from benchmark datasets to practical farming environments, while providing a pathway towards lightweight strawberry disease recognition in settings where extensive annotation and continuous connectivity may not be feasible.
Reproduction assets foundThe paper's own SmartBerry field strawberry image dataset is explicitly stated to be publicly available on Zenodo with a Kaggle mirror. PlantVillage is a cited third-party benchmark, and no author analysis code or trained model checkpoints are stated as available.Dataset · publicip with the SmartBerry partners.
Ethics statement
NA
Data availability
The PlantVillage dataset is publicly available from its
cited source. The SmartBerry dataset is publicly avail-
able through Zenodo, with a Kaggle mirror provided for
convenient access.SmartBerry dataset: Zenodo: https://
zenodo.org/records/22056372; Kaggle: https://www.kaggle.com/datasets/isiborihianle/smartberry-strawberry-field-image-dataset
Funding
This research was supported by the SmartBerry Project
(Project Number: 10071867) and the Department of Com-
puter Science, Nottingham Trent University.
Acknowledgements
The authors thank the Ashley Strawberry Farm staff who
supported image collection, annotation and field eOpen asset ↗Kaggle · isiborihianle/smartberry-strawberry-field-image-datasetpdf-raw-page:17 lines:1-89Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Published9 Sept 2026Journal of Applied Science and Technology Trends
This study proposes an integrated hybrid pipeline approach which is the combination of an optimized Inception-v4 Convolutional Neural Network (CNN), YOLOv8 (You Only Look Once version 8) and extreme gradient boosting (XG Boost) models to detect and classify Tomato plant diseases effectively and accurately. The proposed methodology is based on the fine-grained localization ability of the YOLOv8 model to accurately localize the affected area of the leaf, multi-scale deep feature extraction ability of the optimized Inceptionv4 CNN, and XG Boost to reduce dimensions and optimize features. The hybrid model combines the Inception-v4 CNN, YOLOv8, and XG Boost models with an average CPU processing time of about 200 milliseconds per image, the model performs better in terms of computational efficiency, robustness, and faster prediction. Experimental tests on Plant Village show that the model outperforms the state of the art in 10 disease categories. Specifically, the integrated hybrid system achieved high precision, recall, F1-score, mean Average Precision (m AP), and its training accuracy was more than 0.98 (up to 0.9969), and the testing accuracy was more than 0.96 (up to 0.9695). The hybrid framework shows good reliability even for diseases that are visually similar, such as bacterial spot, early blight, late blight, mosaic virus and yellow leaf curl virus. The optimization of batch size to 32, along with the tuning of the learning rate further improved the stability of training, convergence speed and the generalization of the overall model. The proposed system holds promise for real-time, scalable, and sustainable precision agriculture, aiming for early disease detection and yield protection. Future work includes incorporating Internet of Things (IoT) edge devices, in field environmental monitoring, multimodal data integration and applying Explainable Artificial Intelligence (EAI) techniques to enhance model interpretability for end-users.
Balem JM, Tan C, Dias NCF, Arnold M, Tran S, Severns PM, Teixeira PJPL, Li C, Yang L.
ArabidopsisThermalLeafTissueGrowth / time-series analysisStress response / tolerancePlant / canopy temperature
Repairing damaged tissues is essential for the survival of all organisms. In plants, tissue injury rapidly triggers defense and repair programs. However, the molecular mechanisms linking early injury cues to the later stage of wound repair remain unclear. Here, we show that wounding of Arabidopsis leaves induces localized low temperature at the injury site, likely caused by evaporative cooling, which is accompanied by an activation of cold-responsive genes. Using thermal imaging combined with computer vision and deep learning, we developed a workflow to monitor the dynamics of wound healing in a quantitative, non-invasive, and real-time manner. Mechanistically, we show that C-repeat Binding Factor (CBF) transcription factors are required for the activation of the injury-associated cold response and downstream salicylic acid (SA) signaling. Our findings suggest that the CBF-SA pathway acts coordinately to promote lignin and callose deposition, thereby facilitating wound repair. Together, these findings reveal a link between a wound-induced biophysical cue and the tissue repair program.
Abstract Accurate estimation of maize leaf nitrogen content is important for improving nitrogen-use efficiency and supporting precision crop management. However, leaf-level hyperspectral modeling is challenged by high spectral redundancy and heterogeneous spectral responses among local leaf regions. This study proposes a Hyperspectral–Region Aggregation Network (HSRAN) for maize leaf nitrogen content estimation from region-level hyperspectral spectra. HSRAN consists of a Spectral Adaptive Recalibration Encoder (SARE) and a Context-Aware Gated Aggregation Module (CAGM). SARE performs band-wise residual recalibration and extracts regional spectral representations, whereas CAGM models contextual dependencies among regional features and performs gated attention-based aggregation for leaf-level prediction. Field experiments were conducted in 2024 and 2025 at the jointing, silking, and maturity stages. HSRAN was evaluated against PLSR, RF, XGBoost, SVR, 1D-CNN, MLP, and Transformer1D models. Across the stage-specific and pooled datasets, HSRAN achieved the highest R² and the lowest RMSE while maintaining competitive MAE values. On the pooled full-growth-period dataset, HSRAN achieved an R² of 0.84, an RMSE of 3.63 g kg⁻¹, and an MAE of 2.59 g kg⁻¹. At the jointing, silking, and maturity stages, the corresponding R² values were 0.56, 0.76, and 0.72, respectively. Ablation experiments indicated that integrating SARE and CAGM improved R² from 0.80 to 0.84. To interpret regional contributions, the learned attention weights were mapped back to the original leaf coordinates recorded during regional sampling. Regions near leaf veins, tips, and margins often received relatively higher attention weights, suggesting that their local spectra provided informative cues for model prediction. These findings indicate that spectral–regional joint modeling can improve leaf-level hyperspectral estimation of maize nitrogen content. HSRAN provides a practical framework for non-destructive nitrogen assessment in maize.
Abstract Jute is one of the most important cash crops in Bangladesh and plays a vital role in the country's economy. However, the yield and quality of jute are often affected by various plant diseases, which are difficult to detect manually at an early stage. Early identification is essential to prevent large-scale damage and ensure sustainable production. This study proposes an automated jute disease detection system using image processing and deep learning techniques. Several models, including a Custom Convolutional Neural Network (CNN), VGG16, DenseNet121, and a Hybrid EfficientNetB7 architecture, were implemented and compared. The proposed Hybrid EfficientNetB7 model achieved the highest classification accuracy of 99.7%, outperforming the other models. The system effectively distinguishes between healthy and diseased jute leaves and stems, providing a reliable tool for early disease detection. This research demonstrates that deep learning-based ensemble and transfer learning methods can significantly improve the accuracy and reliability of plant disease detection systems. The findings can contribute to the development of intelligent agricultural tools, empowering farmers with faster and more accurate disease diagnosis for better crop management and yield improvement.
Abstract Tomato leaf diseases (TLDs) such as Early Blight (EB), Late Blight (LB), and Leaf Mold (LM) have a negative impact on crop yield and quality, result-ing in economic losses in agriculture. Traditional diagnosis methods depend on the visual recognition of a disease, which are time-consuming, subjective and may be difficult to reach in rural settings. Keeping these drawbacks in mind, the present work introduces a Transfer Learning model for tomato leaf disease classification based on EfficientNetB3 network. A pretrained Ef-ficientNetB3 network, trained on ImageNet, is used to obtain discriminative features like lesion boundaries, discoloration, fungal textures, and infection spots. Images are cropped to 224 × 224 pixels and split into training, validation and test set. The proposed architecture employs Global Average Pooling, a dense layer with ReLU activation, drop out regularization and Softmax classifier for classification of tomato leaf images into four classes: Early Blight, Late Blight, Leaf Mold and Healthy. Experimental results show the high classification accuracy and low training, validation and test-ing losses. Inter-class misclassifications of the confusion matrix show very few, which supports good generalization. Moreover, Grad-CAM visualiza-tions can generate interpretable heat maps that point to the regions of the image affected by the disease, and multi-class ROC analysis gives high AUC values, which means that it has excellent class separability. The proposed framework provides a precise, reliable, and interpretable approach for auto-mated tomato leaf disease diagnosis, contributing to precision agriculture by assisting in early detection and prompt management of tomato diseases.
Reproduction assets foundThe paper's tomato leaf disease classification uses a publicly available Kaggle dataset (PlantVillage-derived tomato leaf images, 4000 images across 4 classes), explicitly declared in the Data Availability statement with a public URL. No author code, trained models, or other paper-specific assets are disclosed.Dataset · publicitted in accordance with the Journal policies.
Permission to use third-party material
Images or figures are never published previously and did not take from any
internet resources.
Data Availability
The dataset used in this study is publicly available from the PlantVillage
tomato leaf disease dataset on Kaggle’s following link:
https://www.kaggle.com/datasets/kaustubhb999/tomatoleaf
Acknowledgements
The authors would like to thank SR University, Warangal, Telangana,
INDIA and Sharda University, Greater Noida, Uttar Pradesh, INDIA for
providing research facilities and computational resources to carry out this
work.Open asset ↗Kaggle · kaustubhb999/tomatoleafpdf-raw-page:20 lines:1-18Plant phenotyping relevance matchOpenAlex · arXiv · checked 15 Sept 2026
Accurate 3D plant organ segmentation is fundamental to automated phenotyping. Existing approaches rely on annotated training data or species-specific model configurations. We present an annotation-free pipeline for 3D plant organ segmentation, combining text-prompted SAM3 segmentation with semantic neural radiance fields (NeRFs). Given only multi-view RGB images and a list of class names, our zero-shot pipeline produces semantically labeled 3D point clouds without manual annotation, per-species fine-tuning, or domain-specific preprocessing. Multi-view NeRF fusion acts as effective implicit consensus mechanism that lifts imperfect per-frame masks into accurate 3D labels. On a controlled Begonia maculata testbed the SAM3 pipeline achieves 92.6% mIoU, reaching 95.9% of the oracle upper bound established with perfect ground-truth masks. The pipeline was further evaluated on a new dataset spanning ten diverse plant point clouds reaching an average 0.856 mIoU, with leaf and pot IoU above 0.91 and 0.90 for every species, respectively. These results demonstrate that annotation-free 3D plant organ segmentation is now feasible and approaching the range of supervised methods.
Medicinal plants are critical sources of bioactive materials, yet efficient field-scale monitoring of their growth responses to water and nitrogen management remains limited. The present work evaluated Unmanned Aerial Vehicle (UAV) multispectral sensing for estimating leaf area index (LAI), underground fresh weight (UFW), aboveground fresh weight (AFW), underground dry weight (UDW), and aboveground dry weight (ADW) in Astragalus membranaceus var. mongholicus (Bunge) P.K.Hsiao ( A. membranaceus ) under 15 irrigation-nitrogen treatments spanning irrigation inputs of 15.0-30.0 mm per event and nitrogen application rates of 0–450 kg N ha⁻¹. UAV imagery and ground measurements were acquired from 45 plots at 60, 75, 90, 105, 120, and 135 days after sowing (DAS). Thirty-seven vegetation indices (VIs) were screened, and random forest (RF), support vector machine (SVM), back-propagation neural network (BP), genetic algorithm-optimized BP neural network (GA-BP), and decision-level fusion (DLF) models were evaluated with repeated random partitions. Growth responses were trait-dependent. W1N5 reached the peak final LAI (1.96), whereas W2N2 achieved the maximum UFW, AFW, UDW, and ADW (4363.94, 7406.00, 1354.50, and 2596.51 kg ha⁻¹, respectively). VI-trait associations changed with growth stage. GNDVI correlated most closely with LAI at 75 DAS (r = 0.78), and biomass-model performance was stronger during 90–120 DAS than at 60 DAS. DLF was retained in 24 of the 30 stage-specific models. In whole-season modeling, LAI was best estimated by BNDVI input with BP (R 2 = 0.96; RMSE = 0.01), whereas UDW was best estimated by DVI input with DLF (R 2 = 0.93; RMSE = 0.39). Stage-adaptive feature selection increased R 2 by up to 38.60%. Overall, the results indicate that UAV multispectral sensing, stage-adaptive VI screening, and ensemble learning provide a practical framework for monitoring A. membranaceus growth and supporting precision irrigation-nitrogen management in medicinal plant production.
Plant diseases, being a subject of interdisciplinary research, significantly reduce crop yield, quality, and economic returns, while the misidentification of pathogens often leads to ineffective treatments and may harm beneficial organisms and ecosystems. This work develops an approach for robust visual classification of plant diseases under limited and heterogeneous data based on multi-scale fractal texture descriptors integrated into a convolutional neural network. The proposed method employs wavelet transform modulus maxima to extract two complementary fractal characteristics, local fractal dimension and singularity spectrum width, from leaf images at several spatial scales. These descriptors form multi-channel fractal maps fed into a fractal attention module (FAM) inserted after the third stage of a ResNet-50 architecture. The FAM learns to emphasize spatial regions where fractal properties are most discriminative, while a parallel branch encodes global fractal statistics into an auxiliary vector combined with backbone features at the final classification layer. Experiments are conducted on a large heterogeneous collection of 11 public plant disease datasets under 5-shot, 50-shot, and full-scale training regimes. The fractal-augmented model raises classification accuracy from 57.06% to 67.73% on 5 shots and from 80.81% to 86.11% on 50 shots, red outperforming the plain ResNet-50 in these settings, converges within 1–2 epochs versus 25–40, and shows markedly better resilience to color distortions, random occlusions, and grayscale conversion in most cases. The generated attention maps provide spatially explicit explanations of the model’s decisions, increasing transparency for practical use. The proposed approach demonstrates that fractal analysis, embedded as a modulating signal inside a deep network, can serve as an efficient and interpretable inductive bias, which is particularly valuable under data scarcity and noisy agricultural imagery.
Abstract The timely detection of plant health status is essential to obtaining several advantages, such as increasing crop yield, reducing the toxic pollutants used in crops, enhancing healthy crops, and enhancing economic returns. A computer-aided plant status identification system enables plant health identification using plant leaves because plant leaves are the most prominent in the plant. Deep learning (DL) and convolutional neural networks (CNN) are distinguished in the area of plant health identification. Still, CNN cannot handle images of variable size and fails to extract features efficiently if the image has complex backgrounds. To avoid these problems, this research proposes a hybrid model for plant (groundnut) illness status identification using the CNN and vision transformer (HCVT). CNN is efficient in identifying local features, and ViTs are efficient in identifying global features from leaf images; leveraging these advantages enhances the model's efficacy. The HCVT model used the two datasets, groundnut and rice datasets, for experimentation. The HCVT model obtained an accuracy of 96.40% on the groundnut leaf image dataset, with 99.86% accuracy on the rice dataset. The ablation analysis performed the necessity of each component in the HCVT model. The LIME (local interpretable model-agnostic explanations) technique is utilized to comprehend the HCVT method functionality and its results showed that the HCVT model accomplished better than the present cutting-edge models in identifying plant illness and exhibiting its generalization potential. The HCVT model is affordable and widely accessible for analysing plant leaf images. Utilizing the HCVT model offers a robust and easily accessible approach for diagnosing plant diseases by analysing leaf images.
Abstract Maize plant plays crucial role not only in the field of agriculture but also in global economy, since it is the third most cultivated crop across the globe. However, these plants are usually affected by various types of diseases such as blight, common rust, gray leaf spot, etc., protecting the plants from these disease is very important. This research proposes a new deep learning model for disease detection to perform well than other models. The hybrid model uses MobileNetV3 as the backbone architecture integrated with attention, fusion and head. The framework includes data preprocessing, feature extraction, classification and interpretation. We will compare the performance of this model with other models such as VGG16, ResNet50, DenseNet121, ALEXNET, etc. criteria for the final evaluation includes Accuracy, Precision, Recall, F1score, Specificity, Logloss, AUC-ROC curve. Through this proposed model we have achieved an accuracy of 98% which is high than the other models compared. The lightweight nature of MobileNetV3 enables us to implement the model in the mobile and IoT devices also. The present study contributes to the development of deep learning model in the field of agriculture, offering a efficient solution for early maize leaf disease detection.
Reproduction assets foundThe paper's maize leaf disease detection model (MAFH) was trained and evaluated entirely on a public Kaggle image dataset, which the authors explicitly declare in the Data Availability statement. No author code, trained model checkpoints, or other paper-specific assets are stated as publicly available.Dataset · publicig and real-
time datasets and in all the environmental situations.
Funding: This research received no external funding.
Disclosure statement: The authors declare no conflict of interest.
Data Availability
The datasets generated and/or analyzed during the current study are available in the CORN OR
MAIZE LEAF DATASET repository,
https://www.kaggle.com/datasets/smaranjitghose/corn-or-maize-leaf-disease-datasetOpen asset ↗Kaggle · corn-or-maize-leaf-disease-datasetpdf-raw-page:27 lines:1-34Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Published7 Sept 2026Microscopy research and technique
Mangrove ecosystems are frequently exposed to high concentrations of iron (Fe) in sediments, resulting in Fe accumulation in plant tissues. Although Fe is an essential micronutrient involved in several metabolic processes, its excess requires efficient mechanisms of compartmentalization and storage to maintain cellular homeostasis. Histochemical detection using the Perls reaction has usually been applied to identify ferric iron (Fe 3+ ) in biological tissues; however, the combination of this technique with ultrastructural and elemental analyses remains relatively unexplored in plant cells. In this study, we investigated Fe localization in leaf tissues of Rhizophora mangle L. (Rhizophoraceae), a dominant mangrove species, by combining complementary approaches, including Perls cytochemical reaction, transmission electron microscopy (TEM), scanning transmission electron microscopy coupled with high-angle annular dark-field imaging (STEM-HAADF), and energy-dispersive X-ray spectroscopy (EDS). Perls-positive electron-dense deposits were visualized at the ultrastructural level, and their elemental composition was further characterized by EDS analyses. Fe-containing deposits were detected in the epidermis, mesophyll parenchyma, mucilage cells, and vascular tissues, as well as in multiple cellular compartments, including plastids, mitochondria, vacuoles, cell walls, intercellular spaces, and plasmodesmata, whereas sclerenchyma cells showed no detectable Fe-containing deposits. The combination of Perls reaction with TEM, STEM-HAADF, and EDS provides a complementary approach for high-resolution visualization and elemental characterization of Fe-containing deposits at the subcellular level. This integrated methodology may facilitate the investigation of Fe distribution and compartmentalization in plant tissues under contrasting conditions of Fe availability.
Rice leaf diseases significantly reduce agricultural yield and pose a major challenge for sustainable food production, particularly owing to the limitations associated with manual and visual inspection methods that are subjective and often ineffective in early-stage detection. This investigation proposes a new Hybrid ResConvolutional Neural Network (HyResCN-Net) for effective classification and severity identification of rice plant leaf diseases. The proposed framework integrates an Internet of Things (IoT) -based data acquisition and routing simulation using CrowWhale Energy Trust Routing (CrowWhale-ETR) for efficient data handling. Initially, preprocessing is done by an averaging filter to reduce noise. Then, plant leaves are segmented using the Eff-UNet++ method. Augmentation techniques like rotation, scaling, and color change are applied to expand the dataset. Features, like entropy with Gradient Directional Pattern (GDP), Complete Local Binary Pattern (CLBP), and histogram features, are extracted to enhance feature representation. These features are then used within the proposed HyResCN-Net model, which integrates Parallel Convolutional Neural Network (PCNN) and ResNeXt to improve discriminative learning for disease classification and severity estimation. Experimental evaluation is conducted on the Rice Leaf Bacterial and Fungal Disease Dataset. Considering a k-value of 8, the HyResCN-Net gains an accuracy of 94.258%, a True Positive Rate (TPR) of 96.479%, a True Negative Rate (TNR) of 92.898%, a precision of 91.312% and an F1-score of 93.824% compared to existing methods. The HyResCN-Net efficiently enhances rice leaf disease identification and severity analysis, supporting its applicability in precision agriculture applications.
Abstract Plant diseases substantially reduce global crop yields, and cotton production is particularly vulnerable to field-acquired variability in symptom appearance, background clutter, and illumination changes that limit the reliability and scalability of expert visual inspection. This study aimed to develop an accurate, computationally efficient, and explainable framework for real-time cotton leaf disease recognition that is suitable for deployment on resource-constrained edge devices. Using the SAR-CLD-2024 dataset (322 RGB images captured under natural agricultural conditions across seven categories, including healthy and diseased leaves), images were preprocessed via resizing and normalization and augmented online in the training set (random rotations, flips, brightness/contrast adjustments, and random cropping). An EfficientNet-B3 backbone initialized with ImageNet-pretrained weights was fine-tuned using categorical cross-entropy loss and Adam optimization, with early stopping, checkpointing, regularization, and a fixed-seed 70/15/15 train–validation–test partition to enhance reproducibility and reduce leakage. Performance was evaluated on an independent test set using accuracy, precision, recall, F1-score, MCC, balanced accuracy, Cohen’s kappa, confusion matrix, multi-class ROC/AUC, and precision–recall analysis, alongside computational benchmarking (parameters, FLOPs, memory, and inference latency) and comparative experiments against contemporary CNN, lightweight, and transformer-based models. The model showed stable convergence over 30 epochs with a small training–validation gap, predominantly correct predictions with limited confusion among visually similar classes, consistently high precision–recall behavior under moderate class imbalance, and stable performance across repeated runs with low variability and a tight confidence interval. Grad-CAM heatmaps localized necrotic lesions, discoloration, and infected tissues while largely ignoring background, and failure cases were associated with early-stage symptoms, occlusion, shadows, and inter-class similarity. Overall, the framework provides a reproducible, interpretable, and efficient solution for cotton leaf disease classification with practical implications for trustworthy, low-latency, on-device decision support in precision agriculture.
Reproduction assets foundThe paper's Data Availability statement explicitly names the SAR-CLD-2024 cotton leaf dataset used for all experiments as publicly available on Kaggle with a direct URL. No author analysis code or trained model checkpoint is deposited.Dataset · publicntribute to the development of fully automated, scalable, and real-time smart agriculture
systems.
Declaration
Funding
Datta Meghe Institute of Higher Education and Research Wardha, Maharashtra, India
Data Availability: The SAR-CLD-2024 cotton leaf dataset used in this study is publicly
available through the Kaggle platform at: https://www.kaggle.com/datasets/pantho12/sar-cld-2024-dataset-for-cotton
This dataset includes annotated images of various cotton leaf diseases collected under diverse
environmental conditions. All data utilized in this work are freely accessible, and the data
processing methodology has been described in detail to facilitate reproducibility.
Conflict of interest The aOpen asset ↗Kaggle · SAR-CLD-2024pdf-raw-page:32 lines:1-38Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
This work presents a gold nanoparticle (Au NP)-boosted disposable paper-based electrochemiluminescence (ECL) biosensor for minimally invasive on-leaf in situ detection of endogenous H 2 O 2 in tomato leaves. The inherent capillary action of filter paper (FP) was employed to simplify reagent delivery, while gold nanoparticles (Au NPs) catalytically activated H 2 O 2 to generate reactive oxygen species, thereby driving the ECL signal output. This simple, low-cost paper-based platform enabled time-resolved monitoring of H 2 O 2 in stressed plants, providing a reliable in situ strategy for evaluating tomato physiology and early disease/pest warning.
Naincy Sagar · Nassim Belmokhtar · Nathalie Boizot · Ana Alves · David Chassagnaud · R Fichot · José Rodrigues · Luc Pâques
Raman / spectroscopyLeafTissuePhysiological trait estimationLeaf traitsWater status / transpiration
Phenotyping extensive populations remains a major constraint in tree breeding programmes, particularly due to the time-consuming and labour-intensive nature of conventional methods. Near infrared (NIR) spectroscopy, which is a high-throughput phenotyping method, offers an alternative solution, providing a rapid and cost-effective approach for assessing growth- and function-based traits on large numbers of trees. This study aimed to evaluate the potential of NIR spectroscopy-based models for predicting such traits in Larch. Specifically, delta carbon-13 ( δ 13 C), carbon (C), nitrogen (N), specific leaf area (SLA), leaf dry matter content (LDM), and phenolics on needles; the branch hydraulic trait (P 50 ), and lignin and hydroxyphenyl/guaiacyl (H/G) ratio on wood cores from an experimental study on Larix species were predicted using multivariate modelling, specifically, partial least squares regression. Reliable models were obtained for N content (R 2 training = 0.95, r 2 testing = 0.94), lignin (R 2 training = 0.95, r 2 testing = 0.94), and H/G ratio (R 2 training = 0.88, r 2 testing = 0.89), while moderate predictive performance was observed for C content (R 2 training = 0.79, r 2 testing = 0.79) and δ 13 C (R 2 training = 0.76, r 2 testing = 0.69). This methodological approach and its results encourage the transition from traditional laboratory methods to efficient, large-scale-based trait evaluation techniques in forestry.
Shail Bala · S. I. Harlapur · Aditya Kamalakar Kanade · M. P. Potdar · Gurupada B. Balol · V. B. Kuligod · Lingareddy Usha Rani
Field / plotLeafClassificationDisease symptoms / severity
Early and accurate detection of plant diseases is critical in precision agriculture to improve crop management and yield. Mungbean ( Vigna radiata L.) is highly susceptible to several foliar diseases, including yellow mosaic, powdery mildew, leaf crinkle, and cercospora leaf spot, which cause substantial productivity losses. Despite expanding applications of deep learning in plant disease diagnosis, systematic multi-architecture evaluation for mungbean disease classification under natural field conditions remains limited. This study addresses this gap by evaluating five state-of-the-art deep convolutional neural network (DCNN) architectures on a large-scale, field-acquired mungbean dataset that captures real-world variability across environmental conditions and disease severity levels, distinguishing it from controlled laboratory studies. A total of 5,617 original images across five classes were used. Data augmentation was applied exclusively to the training subset after stratified splitting to prevent data leakage. The dataset was partitioned into training (70%), validation (15%), and testing (15%) subsets. VGG16, VGG19, ResNet50V2, DenseNet121, and InceptionV3 were evaluated using identical transfer learning and fine-tuning protocols. Model performance was assessed using AUC-ROC, Cohen's kappa coefficient, McNemar's test for pairwise statistical comparisons, five-fold cross-validation, and Grad-CAM-based interpretability. On the independent test set, InceptionV3 achieved the highest accuracy (98.47%) and macro-F1 (98.49%), followed by VGG16 (98.36%) and VGG19 (97.89%). AUC-ROC values exceeded 0.997 for all models, confirming excellent class discrimination. Grad-CAM visualizations further confirmed that model predictions were based on biologically relevant disease symptoms. The findings demonstrate the effectiveness of deep learning for robust disease recognition under realistic field conditions and highlight the potential of AI-based diagnostic tools for crop health monitoring, precision agriculture, and decision-support systems in mungbean production.
Leaf hyperspectral reflectance can provide a scalable way to estimate photosynthetic capacity (Vcmax), but models trained in one species or measurement context often lose accuracy in another. This transfer problem limits the use of spectral approaches in multi-species crop phenotyping and carbon-cycle applications. Here, we tested physiology-informed inputs for leaf-level Vcmax25 retrieval using paired gas-exchange and reflectance data from wheat (C₃; n = 198) and maize (C₄; n = 81) grown under contrasting nitrogen supply. The four input configurations were raw spectra (Mod1), spectra scaled by a PPFD–absorptance proxy (Mod2), scaled spectra augmented with radiative-transfer-derived traits (Mod3), and scaled spectra augmented with a spectral coordination proxy (Mod4). Within datasets, the best models reached R² = 0.82 in wheat, 0.41 in maize, and 0.76 in the combined dataset. In a matched comparison with a common random-forest learner, the spectral coordination proxy Mod4 improved accuracy only slightly over Mod2 in wheat (RMSE −0.51%; p = 0.0058) and maize (RMSE −1.26%; p = 0.0011) but not in the combined dataset (RMSE −0.15%; p = 0.074), and the trait-based Mod3 showed no consistent benefit. When wheat models were tested on measurement dates not used in training, accuracy remained moderate (R² = 0.563; RMSE = 15.07 µmol m⁻² s⁻¹). Despite this within-dataset performance, models applied to the other species without calibration failed in both directions (negative R²), and adding source-species data did not improve prediction even when a few samples of the new species were used for calibration. These results show that physiology-informed input design provides at most small within-dataset gains, and that reliable prediction across C₃ and C₄ crops requires calibration data from the target crop.
Fei Liu · Yingjie Fan · Mingtao Zhou · Huabing Liu · Shuya Chen · Bin Wen
LeafSegmentation
Abstract Accurate leaf instance segmentation is fundamental to quantifying morphological and structural traits in image-based plant phenotyping. However, substantial variations in leaf scale, shape, and orientation, together with dense overlap and occlusion, often lead to ambiguous instance boundaries. In addition, repeated downsampling and feature reconstruction can progressively erode fine structural details, hindering contour preservation and the separation of adjacent leaves. To address these interrelated challenges, we propose DMSHA-Net, a dynamic multi-scale feature fusion and hierarchical attention network for leaf instance segmentation. Its direction-aware Multi-Scale Feature Aggregation (MSFA) encoder captures complementary horizontal and vertical contextual information across multiple receptive-field scales, thereby improving the representation of diverse leaf morphologies. The Dense Feature Aggregation (DFA) decoder selectively integrates deep semantic and shallow structural features through stage-specific attention mechanisms, where self-attention at coarse resolutions models long-range dependencies, whereas lightweight channel recalibration at high resolutions refines local structures and boundaries. The Learnable Feature Fusion (LFF) module subsequently integrates multi-level semantic and boundary features using normalized learned weights. DMSHA-Net achieves Best Dice (BD) scores of 93.17%, 85.12%, and 90.91% on KOMATSUNA, MSU-PID, and the CVPPP-A1 subset, respectively, with corresponding foreground–background Dice (FBD) scores of 98.19%, 91.02%, and 98.24%. These results indicate that DMSHA-Net achieves competitive segmentation performance across the three datasets, particularly in scenarios involving pronounced scale variation and dense leaf overlap.
In situ monitoring of plant responses to stress is one of the most challenging aspects of precision agriculture, and the dynamic control of crop growth according to fluctuating environmental factors. Although fluorescence imaging provides a nondestructive approach for monitoring stress-related biomarkers, its performance is often hindered by the low abundance of endogenous signaling molecules and strong tissue autofluorescence. Here, we report a microneedle-integrated sensing platform that enables sensitive detection of endogenous hydrogen peroxide (H 2 O 2 ) in living plants. The platform incorporates a second near-infrared fluorescent nanoprobe composed of Er 3+ -doped lanthanide nanoparticles emitting at 1550 nm and Mo-doped polymetallic oxomolybdates serving as the H 2 O 2 -responsive unit. Embedding the nanoprobe into custom-fabricated microneedles allows precise positioning on plant midribs for continuous monitoring of H 2 O 2 dynamics. Under stress conditions, the system successfully visualized spatiotemporal fluctuations of H 2 O 2 in living tomato, spinach, and tobacco plants. This work establishes a strategy for early stress diagnosis and developing universal plant health monitoring technologies.
Abstract Maize is the staple crop for millions of people in Sub-Saharan Africa, particularly for Zambia. Unfortunately, maize crops are exposed to several serious threats due to their susceptibility to foliar diseases like Maize Rust, Leaf Blight, Leaf Spot, Maize Streak Virus, and Maize Lethal Necrosis that may lead to great yield losses. Conventional methods of crop disease identification consist of field surveys that are not only subjective but also difficult to conduct for smallholder farmers. This paper presents the design and evaluation of a highly optimized version of the EfficientNet-B0 Convolutional Neural Network for the automatic detection of maize leaf diseases using maize leaf images obtained from real-world scenarios. The proposed model utilized the concept of transfer learning with ImageNet pre-trained weights and was trained on the Mendeley Maize Crop Disease (Leaf) Dataset which consists of 30,120 images in nine maize disease classes. The developed fine-tuned EfficientNet-B0 yielded 97.57% classification accuracy, macro precision of 97.61%, macro recall of 97.64%, and macro F1-score of 97.61%. From these results, it is evident that transfer learning and fine-tuning greatly boost maize disease classification accuracy while ensuring high computational efficiency. This study makes a significant contribution to precision agriculture as it offers an accurate and computationally efficient AI-based maize disease classification model, which could help smallholder farmers in early maize disease classification.
Reproduction assets foundThe paper's sole qualifying asset is the public Mendeley Maize Crop Disease (Leaf) Dataset of maize leaf images used for all phenotyping/classification measurements, explicitly declared publicly available with a URL. No author analysis code, trained model checkpoints, or other paper-specific assets are disclosed.Dataset · publiconflicts of interest to publish the paper.
Consent to Publish
All authors have read and approved the final version of the manuscript and agree to its
submission to Discover Networks.
Consent to Participate
Not applicable
Data Availability
The Mendeley Maize Crop Disease (Leaf) Dataset used in this study is publicly available at
https://data.mendeley.com/datasets/6w6gsvghfw
Clinical Trial Number
Not applicable.
Ethics Declaration:
Not applicable.
Competing interests
All authors declare no competing interests.Open asset ↗Mendeley · 6w6gsvghfwpdf-raw-page:55 lines:1-22Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Brown Rot (Monilinia spp.) and Leaf Curl (Taphrina deformans) are principal fungal diseases affecting peach (Prunus persica L. Batsch) production, lacking validated AI-based diagnostic tools in tropical highland orchards. This study presents a compact Convolutional Neural Network (3658 trainable parameters), applied identically to fruit and leaf classification, integrated with a background-removal preprocessing pipeline and evaluated through stratified 5-fold cross-validation on 800 in-situ images from four orchards in Cómbita and Choachí, Colombia. The proposed architecture achieved mean accuracies of 82.8% (fruit) and 95.3% (leaf), with AUC values of 0.87 and 0.98, and a trained model footprint of approximately 100 KB, supporting storage- and bandwidth-efficient deployment. Benchmarked against ImageNet-pretrained MobileNetV3-Small and MobileNetV2 under an identical protocol, the proposed architecture matched or exceeded MobileNetV3-Small on leaf classification despite a 257-fold smaller parameter count, and achieved comparable or lower inference latency than both larger backbones. To our knowledge, this is the first validated system for simultaneous detection of both pathogens in Prunus persica under real field conditions, combining a compact, deployment-ready architecture with an ablation-verified preprocessing pipeline. The proposed model was deployed in the DurAPP web platform, giving peach growers in tropical highland regions a practical, low-footprint diagnostic tool suited to smallholder farming conditions.
Urban air pollution may alter plant metabolism long before visible damage becomes apparent. Raman spectroscopy was evaluated as a rapid, non-destructive approach to resolve these biochemical adjustments. Mature Quercus ilex L. trees were sampled along a well-defined pollution gradient in Tuscany (Italy), spanning high, intermediate, and low levels of NO₂ and PM₁₀. Leaf Raman spectra revealed coordinated modulation of primary and secondary metabolism. Pigment-related bands (chlorophylls and carotenoids) increased toward the most polluted site, while inducible flavonoid signals showed site-dependent variation consistent with oxidative pressure in superficial tissues. These patterns were consistent with destructive biochemical analyses and chlorophyll fluorescence measurements, which indicated acclimation rather than photoinhibition damage. A composite Raman index showed a close site-level association with NO₂ exposure, suggesting that nitrogen-related urban pollution was the main exposure component linked to the observed metabolic response. Overall, Raman spectroscopy captures the chronic metabolic imprint of urban air pollution in Q. ilex, resolving coordinated pigment reinforcement and defensive activation without sample destruction. This approach provides a rapid and scalable framework for linking atmospheric chemistry to plant functional status in biomonitoring applications.
Accurate and efficient monitoring of tea plant growth parameters via remote sensing is essential for precision plantation management. However, spectral indices relying solely on reflectance often exhibit limited sensitivity in capturing complex tea canopy characteristics. This study developed a data-driven framework integrating spectral reflectance, frequency-domain harmonic components, and spatial texture features to construct tri-feature fusion indices (TFIs) and establish machine learning and deep learning models for tea growth monitoring. Ten-band multispectral imagery was acquired using a UAV alongside synchronous field measurements of leaf and plant biomass and nitrogen accumulation. TFIs were constructed through exhaustive feature combinations and optimized via a data-driven search strategy. Subsequently, random forest (RF), multilayer perceptron (MLP), convolutional neural network (CNN), and transformer models were evaluated using a leave-one-site-out cross-validation (LOSO-CV) strategy. The selected TFIs showed strong associations with tea growth parameters within the investigated dataset, with R2 values up to 0.63 and 0.62 for leaf dry matter and leaf nitrogen accumulation, respectively. Models incorporating selected TFIs achieved cross-validated R2 values of 0.56 for leaf dry matter (MLP), 0.59 for plant dry matter (MLP), 0.73 for leaf nitrogen accumulation (MLP), and 0.68 for plant nitrogen accumulation (CNN). These models exhibited competitive predictive performance comparable to RF, although no statistically significant differences in mean absolute error were observed under site-held-out evaluation. Furthermore, model-derived spatial maps provided insights into fine-scale spatial heterogeneity and potential interannual variations in tea growth parameters across representative plantations from 2024 to 2025. Overall, this study provides a UAV-based framework for tea growth parameter estimation by integrating multi-domain information without requiring additional environmental observations.
Fruit quality is a critical determinant of economic returns in pear production, and maintaining an appropriate fruit load (FL) is essential for achieving high yield and quality. As a direct indicator of canopy photosynthetic capacity and assimilate supply, leaf number constitutes the key biological basis of reasonable FL determination under the leaf-to-fruit ratio concept. However, accurate and efficient estimation of leaf number in mature pear trees remains technically challenging, limiting its practical use in precision FL regulation. Here, we propose a data-driven framework for leaf number and reasonable FL estimation by integrating 3D point cloud-derived canopy structure with machine learning. A pipeline for extracting 3D architectural traits was developed and implemented in the software tool FTPCT, enabling rapid and standardized trait acquisition. Through correlation analysis, multicollinearity diagnosis, and variance inflation factor screening, five key traits strongly associated with leaf number were identified and incorporated into five machine learning models optimized using Bayesian optimization. Among them, the optimized random forest regression model achieved the highest and most stable performance, with R 2 of 0.85, RMSE of 239.74, and MAE of 149.26 for test dataset. SHAP analysis identified tree crown volume as the dominant contributor to leaf number estimation. Field validation demonstrated that FL regulation guided by the proposed framework significantly improved fruit weight and size without reducing yield compared with conventional practices. Notably, the proposed approach avoids explicit leaf-level reconstruction and relies on less canopy-scale traits, substantially reducing data requirements and computational cost, and thereby offering strong potential for rapid, field-deployable FL regulation in large-scale orchards.
Reproduction assets foundThe paper's phenotyping analysis assets are the authors' publicly released LeafNumPred source code and trained models, and the FTPCT software for 3D trait extraction from pear tree point clouds. Phenotype/point-cloud datasets are only available on request.Code · public. Supplementary data
The following is the Supplementary data to this article:
Multimedia component 1
mmc1.docx (1.6MB, docx)
Data availability
Data will be made available on request. Anyone who wants to obtain other public data can contact us at taost@njau.edu.cn. The source codes and models have been made publicly available at https://github.com/Zhang-Fanhang/LeafNumPred, and the FTPCT software has been released at https://github.com/Zhang-Fanhang/FTPCT/tree/Installation-package.
References
1.Tao S., Khanizadeh S., Zhang H., Zhang S. Anatomy, ultrastructure and lignin distribution of stone cells in two Pyrus species. Plant Sci. 2009;176:413–419. [Google Scholar]
2.Zhang F., Wang Q., Yuan K.Open asset ↗Zhang-Fanhang/LeafNumPredhtml-lines:284-315Code · publical variations [34,35]. The method for calculating these traits are shown in the Supplementary information 1.
2.5. Software implementation for 3D trait extraction (FTPCT)
To facilitate efficient and standardized extraction of canopy structural traits from point cloud data, we used a standalone software tool, FTPCT (available at: https://github.com/Zhang-Fanhang/FTPCT/tree/Installation-package), which integrates the trait extraction procedures applied in this study. The software provides a graphical user interface, enabling users to process tree-level point cloud data and extract key 3D structural traits without requiring advanced programming skills.
FTPCT implements a series of predefined proOpen asset ↗Zhang-Fanhang/FTPCThtml-lines:138-149Code / dataset availability confirmedCrossref · checked 15 Sept 2026
Automated plant disease classification from leaf images demands models that jointly achieve high accuracy and efficient training convergence. Standard deep learning approaches process images through a single feature pathway, limiting their ability to capture diverse visual manifestations such as color changes, texture patterns, and spatial distributions. This paper introduces DMCNNA-FBVL, a framework integrating two complementary innovations: (1) a Deep Multi-Component Neural Network Architecture (DMCNNA) employing three specialized pathways processing color-space statistics, texture descriptors, and raw image features before fusing them through SoftMax-weighted aggregation; and (2) Feedback-Based Validation Learning (FBVL), a training strategy that periodically blends validation-set gradients into weight updates to accelerate convergence. Experiments on the New Plant Diseases Dataset (87,000 images, 38 classes) show that DMCNNA-FBVL achieves 98.7% accuracy, 98.8% precision, 98.6% recall, and 98.7% F1-score, outperforming ResNet-50 by 3.2 percentage points ( 𝑝 < 0 . 0 0 1 ). The primary reported metrics are computed exclusively on an independent 10% hold-out test set, whereas the separate 10% validation partition is used during training for FBVL gradient blending and does not contribute to final test evaluation. Five-fold cross-validation confirms stability (98.7% ± 0.15%). Ablation experiments confirm additive gains, while FBVL reduces wall-clock training time by 15% through faster convergence.
Reproduction assets foundThe paper's plant disease classification experiments use the New Plant Diseases Dataset, which the authors state is publicly available on Kaggle. The authors' code and trained models are only promised 'upon acceptance' (no public URL), so they do not qualify as public assets.Dataset · publicral
monitoring systems.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have
appeared to influence the work reported in this paper.
Data availability
The New Plant Diseases Dataset used in this study is publicly available on Kaggle (https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset). Code and trained models will be made available upon
acceptance.
Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit
sectors.
Acknowledgments
[Removed for double-blind review.]
CRediT authorship contribution statement
[Removed fOpen asset ↗Kaggle · new-plant-diseases-datasetpdf-raw-page:19 lines:1-59Plant phenotyping relevance matchEurope PMC · checked 14 Sept 2026
Abstract In Bangladesh, the potato (Solanum tuberosum L.) stands as an indispensable food and cash crop, deeply intertwined with national food security, rural livelihoods, and the broader agricultural economy. However, foliar diseases such as early blight and late blight frequently precipitate substantial yield losses and quality degradation when not identified and mitigated during the nascent stages of infection. Contemporary diagnostic paradigms remain predominantly manual and visual, relying heavily on agricultural professionals, which is often inefficient and inaccessible for remote farmers. While deep learning has demonstrated remarkable efficacy in automated plant disease recognition, existing methodologies frequently lack interpretability, disease severity quantification, and real-world field applicability. This paper introduces a comprehensive, interpretable deep learning-based framework utilizing EfficientNetV2-B0 for classifying potato leaf images into healthy, early blight, and late blight categories. By integrating Gradient-Weighted Class Activation Mapping (Grad-CAM), the model achieves high transparency, highlighting critical prediction regions. Furthermore, a severity assessment module estimates infection percentages, providing actionable treatment recommendations, ultimately enhancing agricultural decision-making.
The automation and standardization of hyperspectral imaging, particularly at high spatial resolution, are essential for advancing plant phenomics and supporting diverse plant science research. We developed HyperBird, a hyperspectral microscopic imaging robot, to automate the acquisition of up to 351 leaf-disc samples in a single tray within approximately 2.4 h and thereby support scalable plant experiments. System calibration established a spatial resolution of 24.3 μ m (full width at half maximum; FWHM), a spectral resolution of 1.78 nm (FWHM), and a depth of field of 4.53 mm, producing hyperspectral data cubes of 2195 × 2000 × 950 pixels across the 400–1000 nm spectral range. System-level and biological-sample consistency assessments confirmed stable spectral measurements during extended scanning sessions and across independently prepared trays. Thermal evaluation confirmed that the 150 W illumination design introduced minimal sample heating during scanning. HyperBird was first evaluated using grape downy and powdery mildews, where no consistent measurable effect of repeated imaging on pathosystem development was detected under the tested experimental conditions. Subsequently, HyperBird was used to characterize spatiotemporal spectral progression in grapevine leaves inoculated with Plasmopara viticola . An automated regional spectral tracing pipeline was developed to isolate spectra from retrospectively traced regions and compare them with whole-leaf averaged spectra across 0–9 days post-inoculation (DPI). Categorical mixed-effects modeling showed that these spatially resolved regional spectra exhibited significant disease-associated spectral change by 5 DPI, with a sharp transition at 6 DPI, whereas whole-leaf spectra showed delayed and weaker disease-aligned changes beginning at 7 DPI. These results demonstrate that HyperBird enables high-throughput, spatially resolved hyperspectral imaging for quantifying localized plant disease progression and provides a scalable platform for studying spectral biology across plant phenotyping applications.
Reproduction assets foundThe paper's Data Availability statement points to a public GitHub repository containing the authors' code and processed data supporting the phenotyping analyses; raw hyperspectral image data are request-only.Code · publicData Availability
857
The code and processed data supporting the findings of this study are available in the GitHub
858
repository at https://github.com/jy773Cornell/HyperBird-Robot. Raw hyperspectral image
859
data are available from the corresponding author upon reasonable request due to file size and storage
860
constraints.
861
Supplementary Materials
862
Supplementary materials accompany this article as a separate document (supplementary.pdf).
863
Supplementary Figure S1. Representative GSAM-based segOpen asset ↗HyperBird-Robotpdf-raw-page:39 lines:1-75Plant phenotyping relevance matchEurope PMC · checked 14 Sept 2026
Wu F, Liu S, Zohner CM, Townsend PA, Crowther TW, Peñuelas J, Yang D, Yang N, Dong T, Xu W, Wang Z, Liu X, Dai G, Dong J, Durán SM, Schneider FD, Zeng Y, Cornelissen JHC, Kattge J, Wu J, Asner GP, Cavender-Bares J, Reich PB, Yan Z.
Global trait axes reveal overarching dimensions of plant functional variation. However, how these dimensions are spatially organized within and across forest types remains unclear. We combined drone-based full-range imaging spectroscopy with crown-level measurements of 16 physiological, morphological and biochemical traits across temperate, subtropical and tropical forests in China to enable spatially-explicit trait mapping. Through site-training scenario, leaf-to-canopy scaling and spectral-domain modelling tests, we find that reliable canopy trait retrieval depends not only on trait and spectral coverage, but also on preserving trait-spectral relationships across sites and scales. Spectral predictions recovered observed multivariate covariation, summarizing crown variation into a leaf-economics dimension and two additional biochemical dimensions related to hydro-thermal regulation and defence/metabolism. Mapping these dimensions revealed distinct community-level trait organization alongside substantial species- and crown-level variation within forests. These findings link remotely sensed trait retrieval to environmental filtering and plant functional differentiation, providing a scalable framework for monitoring forest functional diversity.
Southern corn leaf blight (SCLB) is caused by the fungal pathogen Bipolaris maydis (syn. Cochliobolus heterostrophus Drechsler) and is a common disease of fall crops of sweet corn. Phenotyping for SCLB resistance is performed through visual scoring, which is subjective and may limit genetic gain for this quantitative trait. As an alternative, we integrated computer vision (CV)-based phenotyping, genome-wide association studies (GWASs), and predictive breeding approaches to dissect the genetic basis of SCLB resistance. We utilized a sweet corn diversity panel with 693 genotypes, for which whole-genome resequencing produced a high-density single-nucleotide polymorphism (SNP) dataset. Broad-sense heritability for visual scoring ranged from 0.44 to 0.73, while CV-based phenotyping produced estimates ranging from 0.56 to 0.73 in multi-environment resistance trials conducted across 5 years and three locations. We performed GWAS using 16,755,210 SNPs and identified 41 associated SNPs. Genomic selection (GS) models on visual scoring phenotypes achieved moderate prediction accuracies under cross-validation of untested genotypes across characterized environments (0.22-0.47) and high prediction accuracies when predicting tested genotypes in uncharacterized environments (0.49-0.68). Using CV-based phenotypes for GS, we observed prediction accuracies of 0.45-0.47 under the untested genotypes in the characterized environments cross-validation scheme and 0.59-0.62 under the tested genotypes in the uncharacterized environments scheme. GS demonstrated reliability for ranking the individuals across a gradient of environments. These findings identify candidate loci and predictive breeding strategies to accelerate the development of resistant sweet corn cultivars.
Reproduction assets foundThe authors state that all datasets (phenotype data) and analysis code (CV phenotyping script, customized GAPIT script) are publicly available in their GitHub repository, which is listed in allowed_urls.Code · publiche images taken for each plot were saved in JPG format and analyzed using a CV method. Here, we refer to the CV method as a custom Python script written using the OpenCV library version 4.5.0, a set of tools for CV (Bradski, 2000 ). The Python script used for leaf CV image analysis is available in our public GitHub repository ( https://github.com/Resende‐Lab/SCLB‐Disease ).
FIGURE 1
Leaf imaging set up with QR‐coded plot IDs (bottom right) and color checker for computer vision phenotyping of southern corn leaf blight disease severity in sweet corn.
In the CT19 environment, a black cloth attached to a wooden board was used as the background. A wooden frame was used to clamp the leaves down toOpen asset ↗Resende‐Lab/SCLB‐Diseaselines:199-209Dataset · publicBLUP and BayesB model implemented in BGLR.
ACKNOWLEDGMENTS
This work was supported by the National Institute of Food and Agriculture USDA‐NIFA2018‐51181‐28419, USDA‐NIFA2019–05410, and USDA‐NIFA 2022–51181‐38333.
DATA AVAILABILITY STATEMENT
All the datasets and codes used in this study are available in the following repository: https://github.com/Resende‐Lab/SCLB‐Disease
REFERENCES
Amadeu , R. R.
,
Cellon , C.
,
Olmstead , J. W.
,
Garcia , A. A. F.
,
Resende , M. F. R.
, &
Muñoz , P. R.
( 2016 ).Open asset ↗Resende‐Lab/SCLB‐Diseaselines:566-596Plant phenotyping relevance matchOpenAlex · Crossref · checked 14 Sept 2026
Abstract Grapevine leaves have dorsiventral anatomy with distinct adaxial (upper) and abaxial (lower) surfaces. Although morphological descriptor lists and ampelographic literature provide information on both the upper and lower side characteristics, in practice, the color traits of the upper side have become the focus of scientific publications. This study introduces the practical application of the recently developed LeafLaminaMap software and the use of trichromatic color indices in grapevine characterization. We aimed to compare colorimetric information on the adaxial and abaxial leaf surfaces as well as to explore the potential of machine learning models in classification. Five statistical descriptors (mean, standard deviation, contrast, energy, and entropy) were calculated for 25 RGB-based color indices on both leaf surfaces of 120 samples collected from four grapevine cultivars (‘Chardonnay’, ‘Pinot noir’, ‘Sauvignon blanc’, and ‘Syrah’). Data was subjected to multivariate statistical analysis and machine learning classifiers. Results showed that the abaxial leaf surface had stronger cultivar-specific color signatures, supporting its suitability for cultivar discrimination. These findings suggest that RGB-based analysis of both adaxial and abaxial leaf surfaces has potential for grapevine cultivar discrimination, offering a new perspective for cost-efficient plant phenotyping.
Whitefly-transmitted begomoviruses cause tomato leaf curl disease (ToLCD). In India, at least 15 begomoviruses are known to cause ToLCD, posing a major challenge to resistance breeding. Although several Ty resistance loci have been introgressed from wild tomato relatives, variable resistance responses are frequently observed, likely due to mixed infections and the absence of a standardized disease scoring system. Moreover, limited knowledge of the infecting begomoviruses in resistant genotypes has hindered the effective use of donor lines in breeding programs. This study evaluated 17 homozygous Ty-gene donor tomato genotypes under natural epiphytotic conditions and identified the associated begomoviruses. To quantify disease severity, a robust disease scoring system was developed using coefficient of infection (CI) by integrating symptom parameters- leaf curling, leaf smalling, stunting, and fruiting, with population-level disease incidence. Field evaluations for two years revealed that genotypes carrying both Ty-2 and Ty-3 loci showed higher resistance, though variability existed among them. Genotypes with Ty-3 alone or Ty-5 + Ty-6 combinations also displayed substantial tolerance, and five genotypes were identified as highly resistant. Molecular indexing revealed frequent mixed infections and identified multiple begomoviruses, including a newly characterized species, tomato leaf curl Ty Pusa virus, alongside tomato leaf curl New Delhi virus, tomato leaf curl Palampur virus, tomato leaf curl Gujarat virus, and tomato leaf curl Joydebpur virus. These findings highlight a shift in begomovirus predominance and possible recombination-driven emergence of new variants. This study provides an integrated framework for evaluating ToLCD resistance and emphasizes the need for continuous reassessment of resistance sources to ensure durable tomato cultivar development.
Salt stress can markedly alter seedling architecture, creating a need for non-destructive three-dimensional (3D) phenotyping methods capable of resolving fine plant structures. However, organ-level segmentation of plant point clouds remains challenging because of leaf overlap, slender stems, ambiguous stem–leaf boundaries, and severe class imbalance. In this study, we developed PTV-SegCo, a task-adapted Point Transformer model for organ segmentation and structural phenotyping of coriander seedlings under salt stress. PTV-SegCo integrates efficient channel attention, gated shallow–deep feature fusion, and a combined cross-entropy–Dice loss to improve representation of fine and minority organ structures. The dataset comprised 60 manually annotated 3D point-cloud samples from 12 cultivation trays repeatedly observed over five acquisition dates under six NaCl concentrations (0, 50, 100, 150, 200, and 250 mmol L −1 ). Because the earliest acquisition represented a particularly challenging developmental stage, these 12 samples were used as a fixed early-stage model-selection set, while samples from the remaining four dates were organized into four date-blocked training–validation configurations. Under this internal model-development protocol, PTV-SegCo achieved mean mAcc and mIoU values of 93.05% and 89.05%, respectively, and showed numerically higher performance than its direct backbone PTV-Seg50. These values should be interpreted as internal comparative results rather than as an unbiased estimate of generalization to unseen cultivation trays, and the present results should not be interpreted as establishing the broad competitiveness of PTV-SegCo against other point-based, convolution-based, graph-based, transformer-based, or plant-specific segmentation architectures. After semantic segmentation, reconstructed scenes were metrically calibrated using the known cultivation-tray dimensions, followed by individual-plant separation and quality control. Four reconstruction-derived structural descriptors—plant height, projected area, voxel occupancy volume, and leaf point ratio—were extracted to characterize temporal structural variation under different NaCl treatments. For treatment-level inference, individual-plant measurements were aggregated within each cultivation tray at each acquisition time, with the cultivation tray treated as the independent experimental unit. Independent manual validation showed close agreement for plant height and projected area, with R 2 values of 0.9969 and 0.986, respectively. Overall, the proposed workflow provides a feasible approach for organ-level segmentation and automated 3D structural analysis of small coriander seedlings under salt stress. The extracted descriptors primarily represent reconstruction-derived spatial characteristics and should not be interpreted as direct indicators of physiological status; voxel occupancy volume and leaf point ratio remain without direct external validation.
Surface phenotyping underpins plant science, preclinical animal research and entomology, yet across all three the measurement is almost always a photograph, which records a projection and not the surface itself. Here we present the Gentschinator3000 , an open structured-light platform that brings high-end metric surface measurement within reach of laboratories with no optics expertise, combining documented open hardware, open reconstruction software and analysis workflows for under 4000 Euro in components. It resolves a planar reference to 45 µm local flatness, registers full rotations to a loop closure of 156 µm, and performs stably across acquisition ranges that we define. Applying one workflow to a leaf before and after desiccation, to murine anatomy and to a spread lepidopteran, we find that projection underestimates surface area by 11 to 41 %. That error grows with the condition under study, with the evaluation scale and with the direction of view, so it can confound phenotype comparisons dramatically. In murine limbs a 15-degree change of viewing direction shifts a projected inter-segment angle by up to 23.2 degrees, while the three-dimensional angle does not move. Projection geometry can therefore contribute as much to a measured phenotype as the biology it is meant to quantify.
Reproduction assets foundThe paper explicitly deposits three public Zenodo records: reconstructed 3D surfaces of all specimens (including the leaf and hop cone phenotyping measurements), the authors' analysis notebooks with derived and per-panel source data, and the reconstruction software with build documentation and working examples. All areDataset · publicData availability
1149
The reconstructed surfaces supporting this study are available at Zenodo under
1150
https://doi.org/10.5281/zenodo.22167250.54
1151
Source data for all graph panels are provided with this paper; for panels showing
1152
rendered surfaces, the underlying reconstructions are in the same record.
1153
1154
Code availability
1155
The analysis notebooks, environment specifications, derived data and per-panel
1156
source data are available at Zenodo under
1Open asset ↗Zenodo · 10.5281/zenodo.22167250pdf-raw-page:35 lines:1-52Code · public1151
Source data for all graph panels are provided with this paper; for panels showing
1152
rendered surfaces, the underlying reconstructions are in the same record.
1153
1154
Code availability
1155
The analysis notebooks, environment specifications, derived data and per-panel
1156
source data are available at Zenodo under
1157
https://doi.org/10.5281/zenodo.22167598.55
1158
The reconstruction software, build documentation and minimal working examples
1159
are available at https://doi.org/10.5281/zenodo.22167471.56
1160
The software and analysis notebooks are released under the MIT licence and the
1161
hardware design files under CERN-OHL-P v2. The visible-light platform described
1162
hereOpen asset ↗Zenodo · 10.5281/zenodo.22167598pdf-raw-page:35 lines:1-52Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published31 Aug 2026International Journal of Advances in Intelligent InformaticsCited by 0 · OpenAlex ↗
Anton Anton · Supriadi Rustad · Guruh Fajar Shidik · Abdul Syukur
Pepper / chilliPotatoTomatoLeafClassificationVisualization / data managementDisease symptoms / severity
Early detection of plant leaf diseases is critical for minimizing crop losses and supporting precision agriculture. While Convolutional Neural Networks (CNNs) have demonstrated high accuracy in image-based diagnosis, conventional architectures may not optimally balance spatial localization and channel-wise feature refinement, particularly in multi-crop classification settings. This study proposes a redundancy-aware dual-attention architecture, termed ATSA-DenseNet, which integrates the spatial branch of the Convolutional Block Attention Module (CBAM-Spatial) with Efficient Channel Attention (ECA) within a DenseNet121 backbone. Unlike prior dual-attention frameworks that retain full CBAM and introduce channel-level redundancy, the proposed design isolates complementary spatial and channel mechanisms to improve representational efficiency without increasing computational complexity. The framework is evaluated on controlled multi-crop PlantVillage-derived datasets comprising tomato, potato, pepper, and maize. Across both 3-crop and 4-crop configuration, ATSA-DenseNet consistently outperforms baseline DenseNet121 and single-attention variants, achieving 99.94% accuracy and 0.9994 macro-F1 on the 4-crop setting while maintaining a lightweight footprint (6.96M parameters, 2.87G FLOPs). Grad-CAM visualizations indicate improved localization of disease-relevant regions compared to the baseline. While results are obtained under controlled imaging conditions, the findings demonstrate that redundancy-aware dual-attention enhances feature discrimination efficiency in multi-class agricultural classification tasks. Future work will extend validation to real-field datasets with natural variability.
Plant diseases greatly affect agricultural production, especially in developing countries, where prompt diagnosis can be quite challenging due to the limited availability of experts in real-time. Deep learning techniques for image analysis is gaining popularity and are increasingly considered an alternative to traditional manual inspection of plants. This research presents the evaluation of plant leaf disease detection system based on a convolutional neural network (CNN) optimized with different nature-inspired algorithms. The backbone model is based on the EfficientNet-B0 pretrained on ImageNet. Therefore, transfer learning is used to adapt the model to an updated PlantVillage dataset. Experiments have been conducted with multiple nature-inspired algorithms to improve generalisation and training efficiency of the prediction model. Different data preparation techniques have been carefully applied to the dataset, creating a unified approach to ensure consistency in the preprocessing pipeline for the training, validation, and testing phases. Our experiments indicate that application of the grey wolf optimizer (GWO) for tuning key hyperparameters of the model, including dropout, learning rates, and weight decay produced the best results, with an accuracy around 99.45%.
Abstract - Plant diseases can significantly affect plant growth, productivity, and overall health. This research presents a mobile-based plant analysis system designed to identify plants, assess their health condition, and analyze visible diseases from plant images. The proposed system allows users to capture an image using a mobile camera or upload an existing image. The image is processed and analyzed using machine learning and deep learning techniques. A Convolutional Neural Network (CNN) can be used to learn visual features such as leaf shape, color, spots, and disease symptoms for plant identification and disease analysis. The system also provides a health assessment and disease severity indication to support users in understanding the condition of a plant. A Flutter-based mobile application provides the user interface, while Python and Flask can be used for image-processing and model-serving tasks. The proposed approach aims to provide a simple and accessible tool for preliminary plant identification, health assessment, and disease analysis. Key Words: plant identification, plant health assessment, disease analysis, CNN, deep learning, Flutter.
All living organisms rely on the movement of ions across cell membranes as the fundamental physical basis of their internal energy and signaling, and plants are no exception. Plants perceive, integrate, and respond to environmental stimuli through electrical signals, classified as action, variation, and system potentials, that are coupled with calcium waves, reactive oxygen species, and hydraulic and hormonal changes to coordinate whole-organism responses despite the absence of a nervous system. Yet most studies characterize these signals using a single feature, such as amplitude or spike duration, in a single tissue, an approach that cannot establish how such signals correspond to the underlying ionic activity, mobility, and structural complexity of the signaling environment, or how this correspondence varies across organs. Here, we correlate plant bioelectrical signals with potential ionic energy flow using a multi-domain framework, combining discrete spike events, continuous waveform properties, spectral composition, and signal complexity applied to leaf, stem, and root recordings from tomato ( Solanum lycopersicum ) exposed to different stimulus. Electrical activity with increased stimulus strength, likely reflecting increased ionic flow, with the root showing the largest response. This suggests plant electrical signaling works as a distributed, ion-based information system, useful for stress monitoring and bio-inspired sensor design.
Visible-near infrared (Vis-NIR) spectroscopy provides rapid crop disease assessment; however, poor model generalizability remains a major limitation when models developed for a specific period are applied to batches collected at different times, primarily due to variations in physicochemical properties such as chlorophyll content, moisture level, surface texture, and tissue structure, which induce shifts in spectral distributions across batches. This study investigates deep domain adaptation to enhance cross-batch transferability for sugarcane disease classification. Two batches of healthy and symptomatic leaves were collected at different times using the same spectrometer. A customized one-dimensional convolutional neural network (1D-CNN) was trained on Batch 1 and adapted to Batch 2 using labelled samples through two strategies: retraining only the fully connected layers or fine-tuning all network parameters. Both strategies achieved 94% accuracy, with precision 0.92, sensitivity 0.97 and specificity 0.98, outperforming the non-adapted model and standard-free calibration transfer methods, namely Correlation Alignment and Transfer Component Analysis. These findings demonstrate that deep domain adaptation substantially improves the robustness and transferability of Vis-NIR classification models across heterogeneous sampling batches.
Background: Faba bean is an important legume crop valued for its nutritional and soil-enriching benefits, yet its productivity is severely affected by foliar diseases. Automated image-based detection using deep learning provides a rapid and reliable approach for early disease identification and improved crop management. Methods: This study developed a machine learning-based automated framework for multi-class classification of Faba bean leaf diseases using transfer learning with the VGG16 convolutional neural network. A dataset of 8,021 RGB images collected under natural field conditions was used, comprising four classes: healthy, rust, gall and chocolate spot. Images were resized to 224 × 224 pixels and normalized prior to training. The pretrained convolutional layers of VGG16 were frozen and a custom classification head with global average pooling and dropout regularization was added. Model performance was evaluated using classification metrics. Result: The proposed model achieved an overall classification accuracy of 92.34% and a macro-averaged F1-score of 0.9227 on the test dataset. Strong classification performance was observed across all disease categories, with particularly high predictive accuracy for healthy and rust classes. The findings demonstrate the effectiveness of transfer learning for plant disease detection and highlight its potential for scalable, automated crop health monitoring in precision agriculture.
Rice leaf diseases pose a significant threat to global food security by reducing crop productivity and causing substantial economic losses. The traditional diagnosis method is manual method, which is low in efficiency, subjective and not suitable for large-scale agricultural monitoring. Despite the advances in automated disease detection using deep learning methods like CNNs, GANs, and transfer learning models, these techniques remain highly computational, not very flexible, and struggle to perform well in different imaging scenarios. Considering these drawbacks, this paper introduces an Explainable Deep Q-Learning based CNN framework which employs CNN-based feature extraction and Deep Q-learning-based adaptive policy optimization for rice leaf disease classification. The proposed model continuously refines the classification actions through reward-based learning, which makes the model more robust in various agricultural imaging environments, in contrast to traditional supervised CNN models that have static classification decisions. The proposed model achieved 98.5% accuracy, 98.52% precision, 98.50% recall, and a 98.51% F1-score, outperforming existing CNN, GAN, reinforcement learning, and transformer-based methods. It also offers a high computational efficiency of 14.2 GFLOPs, 248 MB memory consumption, ~ 48 min of training time, and 6.8 ms inference time per image suitable for resource constrained applications in agriculture. The results demonstrate the effectiveness, scalability, and practical applicability of the proposed framework. The proposed framework performs well on benchmark datasets but more research in the deployment of the edge-devices under different real-world agricultural settings will be investigated in future work.
Reproduction assets foundThe paper trains its Deep Q-CNN rice leaf disease classifier on public Kaggle rice leaf image datasets, which are cited with explicit public URLs and qualify as paper-specific phenotyping image inputs. The authors' own derived data/analysis artifacts are only available upon request, so no authors' code or trained modelDataset · publicSoni Gautam. Rice Leaf Bacterial and Fungal Disease Dataset. Kaggle. Available:Open asset ↗Kagglepdf-page:24 lines:1-94Code / dataset availability confirmedCrossref · checked 15 Sept 2026
Plant diseases are still posing a challenge to the productivity, quality of crops, and food security, especially in locations where field diagnosis is based on manual visual inspec-tion. This paper assesses deep learning network-based automated classification of plant leaf diseases on public RGB leaf-image datasets, such as the Kaggle New Plant Diseases Dataset (Augmented) and PlantVillage images. They investigated four archi-tectures: EfficientNetV2B0, ResNet152V2, DenseNet201, and one hybrid Vision Trans-former (ViT)-based model. The steps of the experiment involved loading the dataset, exploratory analysis, preprocessing, resizing, normalizing, augmentation, transfer learning, independent model training, and evaluation metrics such as accuracy, preci-sion, recall, F1-score, training curves, testing results, and confusion matrices. The hy-brid ViT-based model was reported to have the best accuracy of 99.5%. On the smaller seven class subset of PlantVillage, EfficientNetV2B0 scored 98.11%. On the 38-class dataset, DenseNet201 improved test accuracy (97.34) and validation classification ac-curacy (around 98). ResNet152V2 scored 97.01 on the 38-class test set. The results demonstrate that CNN and transformer-based models can help to recognize plant diseases accurately whereas hybrid attention-based structures provide a promising path to enhance fine-grained classification. Since the model notebooks had varying class settings and splits, the comparison is seen as a model-structured assessment as opposed to a precisely identical benchmark across all architectures.
Reproduction assets foundThe paper's phenotyping inputs are two public plant leaf-image datasets explicitly named in the Data Availability statement: the Kaggle New Plant Diseases Dataset (Augmented) and the PlantVillage dataset, both with public URLs. No author code, models, or supplementary materials are deposited (supplementary materials: 'Dataset · publicy available. Plant leaf images were
obtained from the Kaggle New Plant Diseases Dataset (Augmented) and PlantVillage
datasets. The datasets contain publicly accessible RGB images of healthy and diseased
plant leaves used for supervised image classification research.
DATASET SOURCES
Kaggle New Plant Diseases Dataset (Augmented):
https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset
PlantVillage Dataset: https://plantvillage.psu.edu/All processed data, experimental configurations, and model implementation details are
described within the manuscript. Additional materials may be made available from the
corresponding author upon reasonable request.
ACKNOWLEDGMENTS
The author acknowOpen asset ↗Kaggle · new-plant-diseases-datasetpdf-raw-page:24 lines:1-23Dataset · publicgmented) and PlantVillage
datasets. The datasets contain publicly accessible RGB images of healthy and diseased
plant leaves used for supervised image classification research.
DATASET SOURCES
Kaggle New Plant Diseases Dataset (Augmented):
https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset
PlantVillage Dataset: https://plantvillage.psu.edu/All processed data, experimental configurations, and model implementation details are
described within the manuscript. Additional materials may be made available from the
corresponding author upon reasonable request.
ACKNOWLEDGMENTS
The author acknowledges Istanbul Aydin University for academic support and research
guidance duringOpen asset ↗PlantVillagepdf-raw-page:24 lines:1-23Plant phenotyping relevance matchOpenAlex · Europe PMC · Crossref · bioRxiv · checked 5 Sept 2026
Abstract Non-invasive, high-throughput phenotyping tools are needed that can identify environmental effects on plant structure and function to diagnose factors responsible for reduced growth in commercial and non-commercial settings. In this study, we explored whether the integration of 3D-multispectral (3D) and 2D-hyperspectral imaging (HSI), aided by machine learning (ML), could be used to identify environmental stress treatments imposed during plant growth. Controlled environment-grown Nicotiana Benthamiana plants were subjected to a range of abiotic treatments – including different growth irradiances, heat treatment and drought stress – with the treatments resulting in differences in shoot height, biomass, leaf area and spectral reflectance. ML models were trained to identify these treatments using morphological and spectral traits measured at 27, 29, 31, and 34 days after sowing (DAS). A 3D-multispectral scanner was used to obtain information on plant height, biomass, and leaf area. A visible and near-infrared (VNIR) HSI camera provided detailed spectral information for deriving spectral indices including the Normalised Difference Vegetation Index (NDVI), Photochemical Reflectance Index (PRI) and Normalized Difference Red Edge (NDRE). Manual measurements provided baseline comparative data. The 3D-multispectral scanner reliably estimated above-ground traits, with high correlations between manual and scanner-derived measurements. The ML models accurately differentiated among environmental stress treatments, with the fused 3D+HSI model achieving the best overall predictive performance across all evaluated metrics compared with models based on either imaging modality alone. Results demonstrated the effectiveness of combining 3D-multispectral and 2D-HSI data with ML analyses for non-destructive, high-throughput phenotyping. The integration of these techniques enabled non-destructive, high-throughput identification of environmental stress treatments imposed during plant growth.
Agricultural production faces significant annual losses due to plant diseases, with economic impacts exceeding 40 million dollars and contributing to acute hunger affecting over 281.6 million people in 2023. The timely and accurate identification of plant diseases through leaf image analysis is crucial to mitigate these losses and ensure global food security. This study proposes a lightweight Convolutional Neural Network model, inspired by the MobileNet architecture, designed to classify various plant leaf diseases efficiently. Leveraging a publicly available dataset, this research focuses on developing a model that balances high performance with computational efficiency, making it suitable for real-world applications in resource-constrained environments. The proposed model, named LDPNet, achieved an outstanding accuracy of 99.62%, alongside precision, recall, F1-Score, and AUC metrics of 99.16%, 99.08%, 99.11%, and 99.99%, respectively. A comprehensive comparative analysis was conducted against MobileNetV2 and a reference model from previous research, highlighting the superior performance of LDPNet in terms of both accuracy and efficiency. The results demonstrate that the proposed architecture not only maintains high classification performance but also significantly reduces the number of parameters, making it a practical and scalable solution for plant disease identification. This study contributes to the growing field of agricultural technology by providing a robust, lightweight, and efficient tool for early disease detection, with the potential to enhance crop management and reduce economic losses in agriculture.
Abstract Background and aims Plant volatile organic compounds (VOCs) change dynamically with plant development and in response to environmental conditions. However, their potential as non-invasive indicators of phenological progression remains poorly explored. In this study, we developed a framework integrating automated VOC sampling, time-resolved VOC profiling, and machine-learning analysis for the non-invasive assessment of plant phenology. Using soybean ( Glycine max (L.) Merr.), we investigated whether development-associated temporal variation in VOC emissions could delineate and predict developmental phases. Methods We collected VOCs daily under controlled environmental conditions from 16 to 43 days after sowing, spanning the transition from vegetative to reproductive stages, using an automated sampling system coupled with thermal desorption-gas chromatograph-mass spectrometer (TD- GC-MS). To characterise temporal changes in VOC profiles associated with phenological progression, we analysed the daily VOC data using a multi-step pipeline combining statistical filtering and similarity-based network analysis. We defined VOC-derived developmental phases from similarity patterns in the VOC profiles, then developed and evaluated machine-learning models to predict these phases. Key results Seven VOCs exhibited distinct phase-dependent dynamics, including green leaf volatiles and monoterpenes showing characteristic temporal changes during phenological progression. Network-based clustering of VOC profiles resolved five developmental phases closely aligned with conventional developmental stages. A machine-learning model predicted these phases from the VOC profiles with high predictive accuracy on independent test data, demonstrating that phenological progression could be quantitatively inferred from VOC emission patterns. Conclusions Our findings support VOC profiling as a reliable and non-invasive approach for assessing phenological progression in soybean. By extracting temporally structured VOC signals, this framework captures developmental information that may be difficult to obtain through visual observation alone, particularly after canopy closure. VOC profiling offers a practical tool for monitoring crop developmental dynamics and has broader potential for plant phenotyping and precision crop management.
Reproduction assets found保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。Supplement · publicThe peak area matrix obtained from the MS- DIAL analysis (Supplementary Dataset S1) was filtered to remove unreliable features.Open asset ↗lines:66-69Plant phenotyping relevance matchOpenAlex · checked 5 Sept 2026
Introduction Leaf shape is a genetically determined crop phenotype, and its accurate classification underpins soybean germplasm assessment and genetic improvement. Manual classification is highly subjective and struggles to distinguish morphologically similar leaves, while mainstream supervised classification demands large labeled datasets and incurs high development costs. Efficient feature frameworks for soybean leaf categorization are still insufficient. Methods In this study, 581 biologically replicated terminal leaflets sampled from 194 soybean varieties were analyzed at the single-leaflet level using traditional morphological indices and novel leaf contour angular features. Unsupervised K-means clustering was used to classify soybean leaflet morphological phenotypes; t-SNE was applied exclusively for dimensional reduction visualization, while Welch's ANOVA combined with Games-Howell post-hoc tests was adopted to detect inter-cluster phenotypic differences. Clustering stability and external consistency against manual visual labeling were further quantified via Adjusted Rand Index to comprehensively verify the reliability of grouping outputs. Results The results revealed no significant difference in leaflet edge complexity (p = 0.41) between two manually divided leaf groups distinguished by overall leaf outline similarity; these two morphologically similar leaf clusters failed to be fully separated even though the first two principal components accounted for 90.2% of total variance. For K-means clustering, k = 3 achieved better overall performance with a Calinski–Harabasz (CH) index of 395.55, Davies–Bouldin (DB) index of 1.03, and silhouette coefficient (SC) of 0.38, compared with k = 4. Nevertheless, the angular feature attained an F-value of 951.62 in driving sample reallocation across clusters, serving as the core indicator for fine subdivision at k = 4. Under k = 4 clustering, all six morphological indices differed significantly among the four groups (p < 0.05). Additionally, the number of cross-clustered samples increased from 66 to 119 as k rose from 3 to 4, with 96.6% of cross-clustering attributed to the leaflet contour angular feature. Discussion This research provides a novel reference and technical support for the automated identification and fine classification of soybean leaf morphology.
Abstract Purpose Rapid and non-destructive detection of pigment and nutrient traits in tomato ( Solanum lycopersicum L.) leaves is essential for precision fertilization and greenhouse management. However, most existing studies focus on individual traits (e.g., chlorophyll or nitrogen) with isolated models, limiting the establishment of robust analytical workflows across growth stages and cultivation conditions. This study systematically evaluated hyperspectral imaging workflows for estimating pigment and nutrient traits in tomato leaves. Methods Hyperspectral images were collected at the stages of flowering-fruiting, ripening and harvest from tomato plants supplied with nitrogen at 0, 210, 300 and 390 kg N ha⁻¹. Total contents of chlorophyll, total nitrogen and nitrate were measured by standard biochemical assays for model calibration and validation. Result After sample partition with four strategies, seven spectral preprocessing methods were evaluated, namely moving average (MA), Savitzky-Golay smoothing (SG), Gaussian filtering (GF), median filtering (MF), normalization, baseline correction and standard normal variate (SNV), with normalization, MA and SNV yielded the best predictive performance for chlorophyll, total nitrogen and nitrate, respectively. For feature wavelength selection, competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) were used with CARS yielding the best performance for total chlorophyll and nitrate prediction, while SPA was optimal for total nitrogen prediction. By application of the above optimal methods, random forest (RF), support vector machine (SVM), eXtreme Gradient Boosting (XGBoost) and convolutional neural network (CNN) models were developed to predict pigment and nutrient indicators. Conclusion The SVM performed best for chlorophyll ( R c ²=0.823, R p ²=0.431) prediction, while the CNN achieved higher accuracy for total nitrogen ( R c ²=0.826, R p ²=0.780) and nitrate ( R c ²=0.851, R p ²=0.753) prediction. Overall, leaf nitrogen-related traits were predicted more reliably than total chlorophyll, for which validation performance remained limited. Impact These findings demonstrate that sample partitioning, spectral preprocessing, wavelength selection and model selection should be optimized for each target trait rather than applied uniformly. This study provides a methodological basis for non-destructive assessment of tomato leaf N status and precision fertilization management.
While most phenotyping platforms rely primarily on image-based measurements, advanced plant characterization requires the integration of active physiological sensing modali- ties such as chlorophyll fluorescence. We present an autonomous robotic platform designed to perform targeted fluorescence measurements on plant leaves. The system combines 3D plant reconstruction, geometric analysis, and motion planning to localize suitable measurement points and generate collision-free trajectories for a robotic manipulator. A dense 3D model of the plant is reconstructed from multi-view data and used to extract candidate leaf surfaces based on orientation, accessibility, and sensing constraints. These targets are then integrated into a task-level planning framework that guides the end-effector to precise contact or near-contact configurations required for point-based fluorescence acquisition. The platform enables automated, repeatable, and spatially resolved physiological measurements that go beyond passive imaging. By tightly coupling perception, geometric reasoning, and manipulation, the proposed system provides a robotics-driven approach to high-resolution plant phenotyping and opens new directions for autonomous agricultural inspection and plant-aware manipulation.
Reproduction assets foundThe paper states its code is publicly available in the authors' SonyCSLParis GitHub repository (Plant3DImager), which implements the phenotyping perception and motion-planning pipeline. The exact full URL is split across a line break in the supplied text, so the verifiable allowed URL prefix is used.Code · public2 The code is available at https://github.com/SonyCSLParis/ 3 See for example at https://www.youtube.com/watch?v=Open asset ↗SonyCSLParis/pdf-page:4 lines:1-61Plant phenotyping relevance matchOpenAlex · Crossref · checked 5 Sept 2026
Agricultural operations such as pepper harvesting, fruit counting, and field phenotyping rely on accurate visual recognition and instance segmentation algorithms. However, pepper fruits in complex field environments often exhibit slender and curved shapes, partial occlusion, ambiguous boundaries, and adhesion between adjacent instances. Existing object detection and instance segmentation methods therefore struggle to obtain complete fruit masks, which adversely affects subsequent fruit counting, contour measurement, and picking-point localization. To improve the instance segmentation accuracy of occluded peppers in complex field scenes, this study proposes OccPepSeg-YOLO, an improved model based on YOLO11n-seg. First, a P2FreqFusion module is introduced to fuse shallow, high-resolution detail features with deep semantic features, thereby enhancing the representation of fruit edges and tip regions. Second, an ASC module is designed to model the directional and scale-related morphological characteristics of pepper fruits, while a BoundaryGate module strengthens responses at occlusion interfaces and boundaries between adjacent instances. Finally, an OccPepSegment multi-scale prototype segmentation head is constructed, and a BDoU loss function is introduced to improve the boundary consistency of instance masks. Experiments on a self-constructed field-pepper instance segmentation dataset showed that OccPepSeg-YOLO achieved M-P, M-R, M-mAP50, and M-mAP50–95 values of 93.87%, 92.09%, 97.17%, and 82.31%, respectively, representing improvements of 5.59, 3.18, 3.83, and 9.52 percentage points over YOLO11n-seg. Further comparisons with representative YOLO-based instance segmentation models, including YOLOv8n-seg, YOLOv9c-seg, YOLO12n-seg, and YOLOv26n-seg, demonstrated that OccPepSeg-YOLO achieved the best overall segmentation performance. In particular, its M-mAP50–95 exceeded the best competing result obtained by YOLOv9c-seg by 8.35 percentage points. Under a unified repeated-inference protocol on an RTX 3090 GPU using FP32 precision, a batch size of 1, and 640 × 640 inputs, OccPepSeg-YOLO achieved a mean inference latency of 15.801 ± 1.238 ms, a P95 latency of 17.323 ms, and a throughput of 63.29 FPS. These results demonstrate that the proposed model can produce more complete pepper instance masks under leaf occlusion, fruit overlap, and complex background conditions, providing technical support for field-pepper recognition, fruit counting, and visual perception by agricultural robots.
ABSTRACT Leaf shape is a fundamental trait of plant ecological strategies, influencing biotic interactions and ecosystem functioning. However, established quantitative metrics fail to capture subtle variations and irregularities, require user-based reference points or are challenging to compare among taxa with broadly different leaf shapes. In addition, established metrics typically conflate (aggregate) leaf edge complexity and macro-shape complexity, despite their independent functional significance and genetic foundations. Here, we introduce an entropy-based framework to quantify two new complexity metrics: edge complexity and macro-shape complexity. Based on three case studies, we show that these metrics outperform aggregate metrics in predicting Quercus robur chemical traits, provide more intuitive interspecific classifications, and strongly align with human perception. In addition, edge and macro-shape complexity show high complementarity, while aggregate metrics are highly redundant and typically strongly related to leaf area. Emerging as the strongest predictor of leaf chemistry and key visual cue for complexity as perceived by humans, the effects of edge complexity highlight the under-appreciated functional significance of leaf margins. Our framework and the proposed entropy-based complexity metrics thus promise to help unlock the potential of growing digital image archives of leaves, including images from herbaria and fossils, and are technically readily applicable to shapes of algae, bacteria, pollen, and beyond. The accompanying package ShapeComplexity enables the broad application of entropy-based metrics, providing a powerful tool to explore how the shape of organisms and biological structures influences ecological strategies, biotic interactions, and ecosystem functioning while tracking spatial and temporal variation.
Reproduction assets foundThe paper's authors publicly release their ShapeComplexity analysis code (Rust) on GitHub, used to compute the paper's leaf edge- and macro-shape complexity metrics. Supplementary data/analysis code are on Dryad, but that URL is not in the allowed list. RMBG is a generic third-party background-removal model, not a phenCode · publicThe complete, open-source Rust-code (The Rust Team, 2025 ) is publicly available on GitHub ( https://github.com/Thornbach/ShapeComplexity ), ensuring transparency and reproducibilityOpen asset ↗Thornbach/ShapeComplexitylines:86-94Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Emerged Technologies include imparted potentiality to generate adequate food to converge ultimatum of society. However, aspects like, climate change, plant disease, refuse at pollinators as well as others are demanding to farmers. The presence of diseases pulls up the maturation of their corresponding species. Plant disease recognition using image processing and machine learning (ML) with data collected from Internet of Thing (IoT) sensors has been receiving greater amount of interest in recent years, however timely disease detection remains a demanding issues. Despite early disease detection though reduces the risk of destruction to plants nevertheless the noise present in the sensor networks (i.e. Internet of Things) while collecting internet scrapped images compromise overall precision and accuracy. To address on this research gap, i.e., addressing noise with timely disease detection, in this work ML based plant disease detection method called, Intersection Histogram and Rotational Invariant Principal Component Regression (IH-RIPCR) is introduced. IH-RIPCR technique is dividing as pre-processing and feature extraction. First with the raw plant dataset obtained as input from Corn or Maize Leaf Disease Dataset, pre-processing is done employing Intersection Histogram based Contrast Enhancement model. Second with the obtained pre-processed contrast enhanced images is subjected as input to ML-based Rotational Invariance Principal Component Regression feature extraction model to extract relevant features pertaining to corn or maize leaf images for disease detection in an accurate and precise manner. Experiments are conducted with corn or maize plant leaf dataset with different existing methods to verify hypothesis potentiality of IH-RIPCR technique. Significance of technique is evaluated based on numerous parameters in terms of PSNR, processing time, precision, recall and accuracy respectively.
Maize diseases affecting leaves, stalks, and ears can substantially reduce yield and quality; therefore, rapid and accurate recognition in complex field environments is important for intelligent agricultural monitoring. To address the large-scale variation, weak fine-grained texture, and strong background interference associated with multi-part maize diseases, this study proposes YOLOv11-MPD (YOLOv11 for Maize Multi-Part Disease Detection), a maize disease detection algorithm based on YOLOv11n. The method jointly improves spatial position awareness, shallow detail preservation, local-context modeling, key semantic-region enhancement, and lightweight detection-head reconstruction. RFCAConv, C3k2_RFCAConv, and Detect_LSDECD are introduced into the baseline network to strengthen directional texture modeling, multi-scale feature aggregation, and detection-head feature representation. FG-RFCAConv, HGD-C3k2, LCA-C3k2, and GRN-BiAttn are further designed for high-frequency differential gated detail compensation, P3 high-resolution detail enhancement, local-context fusion, and global-response-normalized attention regulation, respectively. Experimental results show that YOLOv11-MPD achieves Precision, Recall, mAP50, and mAP50-95 of 72.3%, 72.8%, 79.5%, and 50.2%, improving YOLOv11n by 2.4, 2.5, 2.9, and 2.4 percentage points, respectively, while reducing parameters from 2.6 M to 2.4 M. These results indicate that, within the scope of the dataset used in this study, YOLOv11-MPD improves multi-part maize disease detection under complex field conditions. However, the current conclusions are limited to the constructed dataset, and further validation using larger multi-region, multi-season, and multi-device datasets is required to evaluate its broader generalization ability.
Abstract Deep convolutional networks now classify leaf images on curated benchmarks such as PlantVillage with accuracies close to the measurement ceiling of those datasets, which has shifted the open questions away from raw accuracy toward two under-reported issues: which components of a composite pipeline actually cause the result, and whether the components that matter on laboratory images are the same ones that matter on field photographs. We address both with a classification framework evaluated under a protocol that fixes every development decision before the independent test data are read. A Mask R-CNN stage localizes the dominant leaf, the accepted box is expanded by a validation-selected 5% margin and resized to a shared 224 by 224 input, and the same localized image is passed to ResNet50, InceptionV3, and MobileNetV2 together with an extractor that produces eighteen colour and shape descriptors. The forty-five class probabilities and eighteen descriptors form a sixty-three-dimensional input to a neural meta-classifier developed by three repetitions of stratified five-fold cross-validation. On a locked 3,101-image PlantVillage test partition of fifteen classes the framework reached 99.77% accuracy and 99.75% macro F1 with seven misclassifications, and a one-component-at-a-time ablation confirmed that every stage contributed. The same design was then trained and evaluated entirely within a separate thirteen-class PlantDoc field-image dataset, where it reached 90.03% accuracy and 89.40% macro F1. This is a within-PlantDoc experiment and not a controlled-to-field transfer test, so the figure measures how the pipeline behaves on field imagery rather than how a PlantVillage-trained model survives a domain shift. The central finding comes from running the identical ablation on both datasets: on clean images the ensemble breadth and descriptors provide the incremental gains, but under field conditions the ordering changes, and leaf localization and learned fusion become the decisive components. Removing localization cost 2.83 accuracy points and replacing the neural fusion with soft voting cost a further 2.08 points, the two largest effects on PlantDoc. The contribution is a controlled and transparent account of where each component of a localization-aware plant disease classifier earns its place, and of how that ordering shifts between laboratory and field acquisition.
Reproduction assets found保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。Dataset · publicThe classification subset used here (20,638 images spanning fifteen pepper-bell, potato, and tomato classes) was obtained from PlantVillage, which is openly accessible at https://www.kaggle.com/datasets/emmarex/plantdisease.Open asset ↗Kaggle · emmarex/plantdiseaselines:314-336Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published26 Aug 2026International Research Journal of Computer ScienceCited by 0 · OpenAlex ↗
Tomato leaf diseases must be identified early and accurately in order to reduce output loss and advance sustainable agriculture. Deep learning models have shown encouraging results in the identification of plant diseases, but their high processing requirements and inability to adjust to field-specific limitations sometimes make it difficult to implement them in real-world applications. For the purpose of accurately and efficiently classifying tomato leaf diseases, we present a convolutional neural network based on AlexNet that is lightweight and field-aware. Our proposed model is designed with less complexity than traditional architectures and is able to perform inference faster without sacrificing accuracy, making it suitable for real-time implementation in low resource agricultural environments. The algorithm was trained and tested on a dataset of 7704 augmented images of tomato leaves from 7 different disease categories. The Lightweight AlexNet achieved accuracy comparable to VGG variants and outperformed deeper models such as ResNet (96.81%) with lower parameter overhead. It was trained from scratch using TensorFlow & Keras on 100×100 pixel inputs, achieving a training accuracy of 99.15% and a validation accuracy of 99.74%. Moreover, an effective structure of the model enables the installation on edge devices, which offers a scalable precision farming solution. Our work helps to bridge the gap between deep learning research and real-world application in agriculture, allowing the development of real-field, resource-efficient disease detection systems.
Plant diseases can reduce crop quality and productivity, making early detection an important aspect of modern agriculture. Recent advances in deep learning, particularly Convolutional Neural Networks (CNN), have shown promising performance in image-based plant disease classification. This study proposes an explainable deep learning approach for multi-class plant disease classification using ResNet50 and EfficientNetB0 combined with Grad-CAM visualization. The experiments were conducted using the PlantVillage dataset consisting of 15 classes of healthy and diseased plant leaves.The research process included image preprocessing, data augmentation, transfer learning, model training, performance evaluation, and explainability analysis. The dataset was divided into training and validation sets with a ratio of 80:20. Model performance was evaluated using accuracy, loss, confusion matrix, precision, recall, and f1-score metrics. Experimental results showed that ResNet50 achieved the best performance with an accuracy of 92% and a validation loss of 0.19, outperforming EfficientNetB0 which obtained 76% accuracy and 0.82 validation loss. The classification report demonstrated that ResNet50 provided more stable and consistent predictions across most disease classes. Furthermore, Grad-CAM visualization successfully highlighted disease-relevant regions such as lesions, discoloration, and damaged leaf areas, improving the interpretability of the CNN model. The findings indicate that the combination of ResNet50 and Grad-CAM is effective for plant disease classification and provides better explainability for deep learning-based agricultural applications.
Abstract Purpose : The system for diagnosing diseases in Solanaceae crops (SolanAPP), including tomatoes, potatoes, peppers, and eggplants, represents a promising tool for supporting decision-making in agricultural fields using AI. This system reconciles the computational intensity of multitasking models with the infrastructural limitations of rural environments, thereby increasing digital literacy. Its architecture is based on two fundamental pillars: (i) autonomous, offline operation for the detection and classification of diseases in Solanaceae crops; (ii) a georeferenced epidemiological surveillance network with agricultural recommendations for crop monitoring. Methods : The core diagnostic process combines crop-specific semantic segmentation and disease classification models exported to TensorFlow Lite, enabling on-device visual inference and pixel-level severity estimation. An optional online layer integrates Groq’s large language model (LLM)-based reasoning and Firebase services to generate structured agronomic explanations and facilitate the creation of georeferenced community reports when a connection is available. Preprocessing and management of the dataset were performed using the Roboflow platform. The mobile app was developed natively in Kotlin. Model performance was rigorously evaluated using accuracy, recall, F1 score, mean IoU, mPA, inference latency, model size, and decision matrix. The optimal model for the crop was selected using Simple Additive Weighting (SAW). Finally, the overall framework quality and usability were evaluated in Cuba through a user validation survey using a 5-point Likert scale and aligned with the ISO/IEC 25010 software quality model. Results : The model that yielded the best results for most crops was DeepLabV3+ with MobileNetV2, which achieved a classification accuracy of over 97\% while operating with lower inference latency. Beyond individual diagnoses, the system incorporates a collaborative georeferencing feature that allows users to share observations and precise geographic coordinates of detected pathologies to facilitate regional epidemiological monitoring. The user satisfaction survey yielded a satisfaction rating of 4.5/5, with users highlighting the importance of offline diagnosis. Conclusion : Plant disease diagnosis using computer vision can support earlier intervention in resource-constrained agricultural settings, but practical deployment requires models that are accurate, lightweight, interpretable, and usable under limited connectivity. SolanAPP, an offline-first Android framework for detecting foliar pathologies in Solanaceae crops, not only establishes a framework for disease identification in complex natural environments but also provides a theoretical and practical foundation for automated agronomic treatment recommendations and community-based crop surveillance. Impact SolanAPP is a free framework that supports the synergy between multitask deep learning for offline disease diagnosis and LLM-driven reasoning for decision-making in the field. Beyond the quantitative metrics obtained from the selected models, the deployment of SolanAPP in rural contexts serves a fundamental socio-technical purpose: it acts as a catalyst for open access, digital literacy, and agronomic decision-making under unfavorable development conditions. It also represents a strong effort to foster a collaborative epidemiological surveillance network in the agricultural sector. Although it faces challenges, such as the use of field images for model training, this framework marks a promising step in the deployment of edge AI, balancing technical accuracy with practical utility.
Abstract Plant diseases represent major constraints on agricultural productivity, often resulting in significant yield losses. This study presents FitoView, a cloud-based mobile decision support system that integrates deep learning with real-time weather forecasting for sustainable plant disease management. Demonstrated through a case study on Cercospora leaf spot in chili pepper, the system employs custom YOLOv8 models for automated disease detection, classification, and pixel-level severity quantification, combined with meteorological data from the OpenMeteo API. The core innovation lies in an integrated decision matrix that considers three dimensions: AI-assessed disease severity, 48-hour climatic risk forecasts, and optimal spraying conditions, generating four contextualized management scenarios with tailored re-evaluation periods (3-10 days). OpenMeteo API validation across four cities in Sergipe demonstrated very strong correlations, with Pearson coefficients (r) of 0.90-0.97 for temperature, 0.81-0.95 for humidity, and 0.92-0.95 for solar radiation, corresponding to R² values of 0.65-0.94. The YOLOv8 object detection model achieved perfect precision (100%) and macro-averaged recall of 89% across all disease classes, with Cercospora leaf spot detection reaching perfect metrics (100% precision, recall, and F1-score). Field validation in Lagarto, Sergipe, confirmed the system’s practical use: it accurately detected Cercospora leaf spot, estimated severity, and, combined with climatic risk, generated recommendations for alternative treatment and short-term re-evaluation. The Progressive Web Application architecture, deployed on a Cloud Platform, ensures accessibility without installation requirements, while the modular design enables scalability to additional crops and diseases, representing a significant advancement toward democratizing AI-powered precision agriculture tools for smallholder farmers in Brazil.
Precision agriculture increasingly requires intelligent systems capable of integrating multimodal sensing with transparent decision support to enable timely and reliable crop management. This study proposes a hybrid intelligent IoT framework integrating environmental monitoring, wearable plant physiological sensing, AI-based pest-monitoring, machine-learning-based prediction of crop physiological stress, and explainable fuzzy rule-based decision support into a unified architecture for crop stress assessment. A novel Physiological Stress Index (PSI) was developed by combining vapor pressure deficit, relative humidity, Delta-T, leaf capacitance, and relative irradiance to provide an interpretable indicator of crop physiological stress. The proposed framework was experimentally validated under real field conditions using a commercial environmental monitoring station, wearable leaf sensors, AI-enabled pest-monitoring devices, and cloud-based analytics. Correlation analysis confirmed strong relationships between PSI and the principal environmental variables (VPD: r = 0.980, Delta-T: r = 0.990, RH: r = −0.961), demonstrating the internal consistency and sensitivity of the proposed index. At the 15 min forecasting horizon, Linear Regression and Gradient Boosting demonstrated virtually identical performance: Gradient Boosting achieved a marginally lower RMSE and higher R2 (RMSE = 0.0273; R2 = 0.9810), whereas Linear Regression achieved a slightly lower MAE (MAE = 0.0186). At the 1 h forecasting horizon, Gradient Boosting achieved the strongest performance (R2 = 0.9034), indicating increasing relevance of nonlinear modelling at longer prediction horizons. The proposed framework demonstrates the feasibility of combining multimodal sensing, machine learning, explainable artificial intelligence, and edge-enabled IoT technologies to support proactive, transparent, and intelligent precision agriculture.
Abstract In modern agriculture, it is essential to identify the early symptoms of plant diseases and to accurately maintain the productivity of the crop and reduce economic losses. Foliar diseases are a special concern in peach because they can cause yield as well as quality if not timely detected. Artificial intelligence, machine learning and deep learning are some of the advanced technologies that are gaining great importance in today's agriculture, especially with the image analysis applications. In this study, a deep learning method for automated detection and classification of three peach leaf diseases and healthy class was presented based on image data. The dataset were taken at different phenological and disease stages under temperate conditions in Kashmir with four classes Healthy, Leaf Curl, Shot hole and Rust. Three convolutional neural networks (CNNs) architectures were applied, VGG-16, ResNet 50 and Xception were trained using transfer learning and Inception-V4 was trained from scratch for a comparative study of the learning strategies. Data augmentation techniques were applied to improve generalization. Results show that all models were able to learn disease specific features well. The result of Inception-V4 was found to be highest with 97.50%, followed by ResNet-50 with 94.49%, VGG-16 with 92.04% and Xception with 75.95%. The results of transfer learning-based architectures were also good and competitive but the best results obtained from the Inception-V4 architecture reveal its capability in modelling complex visual patterns. The results highlight the potential of deep learning techniques for early detection of diseases in peach, supporting precision agriculture and better disease management.
Potato, one of the world's most important staple food crops, is highly susceptible to various foliar diseases that significantly affect its productivity and pose a serious threat to food security, thereby contributing to economic losses and impacting farmers’ income. Therefore, early and accurate detection is essential. Conventional detection methods rely primarily on manual observation, which is time-consuming and requires specialized personnel. In this study, five convolutional neural network (CNN) architectures were evaluated for the automatic classification of potato leaf diseases, including pretrained models (ResNet50, MobileNet, and VGG16) and models trained from scratch (AlexNet and LeNet-5). The dataset was constructed by integrating and selecting images from publicly available Kaggle repositories, resulting in a total of 6,691 images distributed across five classes: early blight, late blight, potato leafroll virus (PLRV), mosaic virus (PVY), and healthy leaves. Multiple experiments were conducted by varying hyperparameters such as batch size, optimizers, and the number of training epochs. The results show that VGG16 achieved the best performance, with an accuracy of 99.87%, outperforming the other architectures. Additionally, a mobile application based on the optimal model was developed for real-time detection. These findings demonstrate the potential of deep learning for intelligent and scalable agricultural diagnostic systems.
Live imaging of plant subcellular structures is key to deciphering the spatiotemporal bases of cellular processes, and their functional impact on growth and morphogenesis at various biological scales. Live imaging of plant cells essentially relies on expression of fluorescent markers labeling cells or subcellular structures of interest. Simultaneous multi-channel imaging of several markers is still not routine practice in plant cell biology, owing to issues linked to genetic or spectral compatibility of markers, differences in expression levels, silencing, toxicity, etc. Here we designed a three-color marker in Arabidopsis thaliana and Capsella rubella , enabling high-resolution live imaging of plant morphogenesis, including labeling of the cell membrane, the nucleus and the microtubule cytoskeleton. Detection of MT arrays involved the development of a MAP4-MBD-based microtubule marker optimized for plant cells. The three- color marker allows visualization of the three-dimensional organization and dynamics of plant microtubules within the intracellular space with unprecedented precision, in various organs including the root and shoot meristems, the leaf, anther, and gynoecium. Our results demonstrate the potential of such single-construct strategy for cell biology studies in plants.
Accurately quantifying the areas of necrotic lesions on plant leaves is essential for evaluating plant-pathogen interactions and disease resistance. Although digital image analysis methods using ImageJ are widely employed, they often require case-specific optimization and may not be readily applicable across different experimental conditions. Furthermore, many studies have used ImageJ for lesion measurement without providing methodological details, which limits reproducibility. Here, we present a simple, step-by-step ImageJ workflow for measuring irregular necrotic lesions using a standard personal computer and mouse. The procedure relies on manual lesion selection using the freehand selection tool, followed by Gaussian smoothing, binarization, and automated particle analysis to extract lesion area measurements. By balancing manual isolation with computational thresholding, this protocol eliminates the need for extensive parameter tuning. This approach provides an accessible, reliable alternative to time-consuming color thresholding methods, thereby improving transparency and reproducibility in lesion quantification. The workflow's reproducibility has been confirmed through both intra-user and inter-user analyses. Key features • Relies on simple manual freehand selection combined with minimal image processing, requiring only a standard computer and mouse without specialized software or advanced training. • Enables accurate quantification of irregular necrotic lesions in conditions where automated thresholding methods require time-consuming optimization. • Provides a fully detailed, reproducible ImageJ workflow addressing common gaps in published methods, facilitating direct implementation. • Demonstrates high reproducibility, validated by intra-user and inter-user statistical analyses, ensuring reliable lesion quantification regardless of the operator.
Accurate quantification of plant disease severity is essential for evaluating host-pathogen interactions and assessing the effectiveness of disease management strategies. Traditional visual scoring methods and manual estimation of infected tissue are widely used but are often subjective and prone to observer bias. Digital image analysis offers an objective alternative by enabling automated identification and quantification of symptomatic plant tissues based on color and spatial characteristics. Here, we present a MATLAB-based image processing protocol for differentiating diseased and healthy plant tissue from digital leaf images. The workflow involves acquisition of standardized leaf images, conversion of RGB images into hue-saturation-value (HSV) color space, segmentation of diseased tissue using defined HSV thresholds, refinement of the segmented mask through morphological operations, and extraction of the whole leaf area. The protocol then calculates the diseased area and total leaf area in pixels and computes the percentage of infected tissue. The method uses MATLAB together with the Image Processing Toolbox and can be implemented using simple scripts. This protocol enables rapid and reproducible quantification of disease severity in plant leaves exhibiting visually distinct symptoms such as necrotic lesions or blight patches. By minimizing observer bias and providing quantitative measurements of infected area, the protocol offers a practical and reproducible approach for plant disease phenotyping and evaluation of disease management strategies across diverse plant-pathogen systems where diseased tissues can be clearly distinguished from healthy tissues under reasonably controlled imaging conditions. Key features • A reproducible MATLAB-based workflow for separating diseased and healthy plant tissue using color-space segmentation. • Applicable to plant diseases where symptomatic tissue contrasts clearly with healthy tissue (necrosis, blight lesions, rot patches). • Requires digital leaf images, MATLAB, and the MATLAB Image Processing Toolbox for image processing and disease quantification. • Enables rapid calculation of diseased leaf area and disease severity using automated pixel-based quantification.
Reproduction assets foundThe protocol explicitly deposits its authors' MATLAB image-processing workflow (HSV segmentation, mask refinement, pixel-based disease quantification) in a public GitHub repository with README instructions and example images.Code · publicGitHub repository containing the MATLAB source code, README file with installation and execution instructions, and representative example image(s): https://github.com/pankajborahmajuli-source/Leaf-Disease-Detection-MATLAB-Code/blob/main/README.mdOpen asset ↗Leaf-Disease-Detection-MATLAB-Codehtml-lines:112-148Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Fluorescence in situ hybridisation (FISH) is a valuable technique for visualising RNA molecules in their native cellular context. Still, its application in plant tissues is often limited by tissue autofluorescence and the lack of optimised protocols. Here, we developed and validated a simple, reproducible FISH workflow to detect Apple scar skin viroid (ASSVd) in cucumber. Systematic optimisation of probe chemistry, tissue selection, and sampling stage significantly improved assay sensitivity and reproducibility. The AZDye594-labelled antisense riboprobe produced higher signal-to-background ratios and lower background fluorescence than fluorescein-labelled probes, enabling reliable detection of ASSVd in vascular-associated tissues. The optimised workflow consistently detected ASSVd in both leaves and stems. High-resolution confocal imaging further revealed predominant nuclear accumulation of ASSVd RNA in infected cells. Together, this study establishes a sensitive and accessible FISH workflow for localisation of ASSVd in cucumber and provides a practical platform for investigating the spatial distribution of viroid and other plant RNA pathogens.
Andrew Trlica · Rachel L. Cook · Matthew J. Sumnall
Field / plotLiDAR / point cloudMultispectral / hyperspectralLeafMorphology / geometry measurementLeaf traits
Canopy Leaf Area Index (CLAI) is a stand attribute containing information on the real-time health and growth potential of managed pine plantations. Current remote sensing techniques for quantifying CLAI rely on simple linear models applied to satellite multispectral imagery, or on techniques based on light detection and ranging (LiDAR) data that are costly and less frequently collected. This study demonstrates a convolutional neural network (CNN) approach to retrieving CLAI from 10 m Sentinel-2 multispectral imagery with a model trained on gridded LiDAR-based CLAI estimates. We demonstrate large gains in accuracy with the CNN compared to traditional linear models based on vegetation indices (e.g., Simple Ratio), but also clear shortfalls in model skill when predicting “blind” in some spatial domains that were completely excluded during model training. Pixel-scale root mean squared error ranged from 0.34 to 0.64 by domain when exposed to CLAI training data from all available spatial domains, but rose to 0.58–1.74 when predicting without prior domain-specific training. Prediction accuracy was consistently lower when applied to completely unobserved USGS LiDAR-based CLAI estimates. Traditional linear models, in contrast, had the advantage of usually lower prediction error across unobserved spatial domains (0.43–1.98), but with lower maximum accuracy. These results demonstrate a potential route for deploying more complex models for LiDAR “mimicry”, e.g., between data acquisitions widely separated in time, but advocate for the development and use of more stable generalized approaches for use in unobserved managed pine stands.
Hazra M, Crowther A, McInerney F, Strömberg CAE, Fabillo M.
LeafSeed / grainClassification
Background and aims Phytolith analysis is widely applied in palaeoecological and archaeological research, but its interpretive strength depends on the availability of robust modern reference collections. This study expands the modern Australian phytolith reference collection by analysing 42 native plant species representing 24 families and 37 genera with emphasis on silicification patterns across major growth forms, including forbs, shrubs, trees, and C3 grasses. Methods Phytoliths were extracted from available plant parts, including leaves, stems, flowers, seeds, seed pods, cones, and roots, depending on sample availability. Morphotypes were identified following ICPN 2.0, with grass silica short cell phytoliths (GSSCPs) further classified by shape and size to examine subfamily-level patterns. Phytolith morphotype percentage data were analysed using Hellinger transformation, PerMANOVA, PCA, LDA, and hierarchical clustering to assess compositional differences among plant growth forms and grass subfamilies. Key results Phytolith production varied strongly among growth forms and plant parts. Grasses were abundant producers, whereas most forbs, shrubs, and trees were trace producers or non-producers. Leaves were the most consistent source of phytoliths, while seeds and seed pods were predominantly non-producers. Grass silica short-cell phytolith (GSSCP) morphotypes showed clear subfamily-level differentiation. Pooideae produced Rondel morphotypes. Danthonoideae produced Rondel as well as wide Bilobate types. Panicoideae and Oryzoideae exhibited a pronounced Bilobate signature, commonly associated with Polylobate and Cross forms. Non-grass taxa (woody, shrubs, and forbs) were dominated by Spheroids, Tracheary elements, Epidermal, and Polygonal sheets and other non-diagnostic forms. Phytolith assemblages differed significantly among plant families, with Poaceae uniquely producing GSSCPs, while non-grass families showed greater overlap in assemblage composition. Conclusion By expanding taxonomic and anatomical coverage, this study strengthens the capabilities of phytoliths in the reconstruction of grasslands and in general paleo vegetation in Australia, especially where other proxies such as pollen are limited.
Reproduction assets foundThe authors explicitly state that the R scripts used for data analysis and figure generation are publicly available on their GitHub repository, which directly reproduces this paper's phytolith statistical analyses (PCA, LDA, PerMANOVA, clustering, plots). Supplementary data files contain the paper's measurements but noCode · publicntification of all plant specimens collected for this
study.
A
14 FUNDING
M
15 Funding for this study was provided by ARC Discovery grant DP210100508 and a Ph.D.
D
16 fellowship (UQGSS) to MH.
TE
17 DATA AVAILABILITY
18 The R scripts used for data analysis and figure generation are publicly available on GitHub
EP
19 repository: https://github.com/Manoshi-sporo/Australian-Phytolith-Reference-Collection.
20 CONFLICTS OF INTEREST
CC
21 The authors declare no competing financial or commercial interests.
A
22 AUTHOR CONTRIBUTIONS
23 MH: writing original draft, conceptualization, software, investigation. AC: Supervision, Writing -
24 Review and editing, FM: Supervision, Writing-Review and editing.Open asset ↗Manoshi-sporo/Australian-Phytolith-Reference-Collectionpdf-layout-page:34 lines:1-87Plant phenotyping relevance matchEurope PMC · checked 5 Sept 2026
To address the challenges posed by large lesion-scale variations, weak boundary cues, and high inter-class similarity in rice leaf disease detection, this study proposes an improved RT-DETRv2-R50-based rice disease detection model, termed CED-RTDETR. First, a Lightweight Contour-Guided Aggregation Backbone (LCGA-Backbone) is constructed. In the PResNet residual blocks, an InceptionDWConv2d-based direction-aware depthwise separable spatial mixing strategy is introduced to capture local, horizontal, and vertical lesion texture patterns with low computational overhead. Meanwhile, a Contour-guided Efficient Global Aggregation Block (CEGA Block) is embedded after the outputs of the C3, C4, and C5 stages. Through contour-difference enhancement, channel shuffle, group-wise efficient global aggregation, and bottleneck channel mixing, the proposed block strengthens the representation of the boundaries of small lesions, weak textures, and contextual semantics. Second, a Multi-scale Evidence Interaction Fusion Neck (MEIF-Neck) is used to perform multi-scale evidence interaction and salient-region competition after cross-scale feature concatenation. A lightweight feature reconstruction process is further implemented using Spatial-Evidence RepNCSPELAN (SE-RepNCSPELAN), which is developed from a YOLO-style feature fusion structure. Finally, a Direction-Amplitude Decoupled Deformable Attention (DAD-DA) mechanism is introduced to decompose sampling offsets into direction rotation residuals and radius gains while incorporating an aspect-ratio-aware geometric compression-restoration strategy, thereby improving the geometric stability of sampling locations during the decoding stage. Experimental results on the constructed rice leaf disease dataset show that CED-RTDETR achieves AP, AP50, and AP75 values of 29.5%, 75.2%, and 17.6%, respectively, outperforming RT-DETRv2-R50 by 4.1, 4.9, and 3.8 percentage points. To further evaluate the model on an additional public benchmark dataset, experiments were also conducted on the public Rice Disease Dataset. On this dataset, CED-RTDETR achieves AP, AP50, and AP75 values of 34.1%, 77.1%, and 23.2%, respectively, improving upon RT-DETRv2-R50 by 4.3, 5.6, and 3.4 percentage points. These results indicate that the proposed method achieves consistent overall performance improvements on both the constructed dataset and the public benchmark dataset.
Functional-structural plant models simulate plant responses to environmental conditions, but their development and evaluation are often limited by the lack of datasets combining detailed architectural and physiological measurements. Here, we present a comprehensive dataset acquired from four oil palm plants ( Elaeis guinnensis) grown under controlled and contrasting climate scenarios. The dataset includes (i) three-dimensional reconstructions of plant architecture derived from terrestrial lidar point clouds, (ii) leaf-level gas exchange measurements used to parameterize photosynthesis and stomatal conductance models, and (iii) continuous plant-scale measurements of CO 2 and H 2 O fluxes obtained in a microcosm under precisely monitored and manipulated environmental conditions (light, temperature, humidity, and CO 2 concentration) across height climate scenarios. By combining detailed structural data with physiological measurements at both leaf and whole-plant scales, this database has been designed to build and evaluate digital twins (or shadows) of plants functioning under controlled conditions. It provides a valuable resource for calibrating biophysical models (light interception and photosynthesis), benchmarking model predictions across scales, and investigating the consistency between leaf-level parameterization and plant-level fluxes. All data and processing workflows are openly available, facilitating reuse for model development, evaluation, and intercomparison in plant and crop modelling communities.
Amanda Ewen · Rodrigo Godoy Mendez · Karar Al-Shanoon · Dawn Omoluabi · Anjana Samarasinghe · M. Alejandra Oviedo-Ludena · Karina Chimbo Huatatoca · Kara Glor · Keiko Nabetani · H. R. Kutcher · Lipu Wang · Ian Stavness · Lingling Jin
Reliable and objective phenotyping is essential for plant breeding programs to characterize genetic variation and accelerate crop improvement. Conventional disease assessment relies on expert visual scoring, which is labor-intensive, subjective, and prone to inter- and intra-rater variability. Although image-based phenotyping methods have been proposed, many require manual intervention, specialized imaging setups, or single time-point measurements, limiting their ability to capture disease progression over time. Here, we present a pipeline for longitudinal plant disease phenotyping that quantifies wheat stripe rust and leaf rust progression from time-series images. The pipeline performs semi-automated leaf and automated pustule segmentation from images acquired in situ , enabling objective disease severity estimation with minimal user intervention and without requiring solid backgrounds or manual leaf manipulation or detachment. By extracting temporal traits, including disease severity trajectories and standardized area under the disease progress curve, the method provides a comprehensive characterization of disease development throughout infection. Association between automated and expert assessments was moderate for stripe rust ( R 2 = 0.58) and strong for leaf rust ( R 2 = 0.85), while expert inter-rater reliability was moderate for both diseases (ICC = 0.675 and 0.800, respectively). The proposed approach establishes a scalable and reproducible framework for longitudinal disease phenotyping in controlled environments, with broad applications in disease resistance screening and crop breeding.
Reproduction assets foundThe paper's Code and Data Availability section explicitly states that software and datasets (the phenotyping pipeline and imaging datasets) are publicly available at the authors' GitHub repository and project website, both of which are in the allowed URL list.Code · publicSoftware and datasets are available at:
https://github.com/USask-BINFO/greenskeye_analysis and https://greenskeye.usask.ca/speedbreeding/ .Open asset ↗USask-BINFO/greenskeye_analysislines:195-225Dataset · publicSoftware and datasets are available at:
https://github.com/USask-BINFO/greenskeye_analysis and https://greenskeye.usask.ca/speedbreeding/ .Open asset ↗lines:195-225Plant phenotyping relevance matchOpenAlex · Crossref · checked 5 Sept 2026
Plant diseases lead to substantial yield losses and pose a persistent threat to global food security, creating an urgent demand for high-throughput, accurate, scalable, and non-destructive disease-monitoring approaches. Remote sensing has emerged as a powerful tool, yet progress in plant disease detection remains fragmented across various disciplines, tasks, sensing methods, and data modalities. This review introduces a hex-view perspective to synthesise remote-sensing-based plant disease detection within a cohesive conceptual framework. Instead of treating sensing technologies, algorithms, and datasets independently, the hex-view incorporates six interconnected dimensions that jointly capture how biological processes, the measurement scale, and data characteristics constrain disease detectability, including when detection is possible and how reliably it can be achieved. The hex-view framework comprises six interconnected dimensions and forms an integrated framework called BTSCAD: (1) Biology (B): plant–pathogen interactions constituting the biological foundation of disease development and expression. (2) Task (T): the diverse disease-detection tasks and their corresponding research objectives. (3) Sensor (S): the sensing modalities that define the data acquisition type and richness of captured information. (4) Condition (C): the environmental conditions, sensing platforms, and spatial scales that shape disease observations and bridge controlled experiments and real-world deployment across leaf, canopy, plot, and regional scales. (5) Algorithm (A): the classical and state-of-the-art data-analysis algorithms used to extract disease-related information from sensor data. (6) Dataset (D): the data sources that underpin model development, evaluation, and generalisability. The hex-view perspective provides a clear framework for interpreting previous research and identifying future research directions. This review lays a structured foundation for developing robust, interpretable, and transferable disease-detection systems, supporting advancements in precision agriculture, high-throughput phenotyping, and sustainable crop production.
Abstract Visual assessments of growing forest nursery plants are time-consuming and often result in a lack of information at a physiological level. There exists a need for health screening in nurseries, that is fast and efficient, to improve overall health monitoring and nursery productivity. Rapid handheld sensors such as rapid thermal devices, leaf porometers and moisture meters, can provide regular information at a physiological level, that can improve the understanding of the impact of stress on young plant cuttings and their decline in health over time. This paper evaluates the utility and reliability of contemporary sensor technologies, to operationally monitor stress phases in juvenile forest plant cuttings during progressive moisture (dry-down) conditions. Furthermore, to assess whether thermal sensors could be used as an indicator, in conjunction with other variables such as soil water content or stomatal conductance, is needed operationally for fast screening during limited planting windows. Near Infra-Red Analysis (NiRA) data was collected to understand detailed plant functions at a finer reflectance level. A relationship was found where the increase in thermal signals reflects a depletion of water content, resulting in an eventual decline in stomatal conductance and, ultimately, plant mortality. Several algorithms were used in a preliminary test, using RapidMiner software, to discriminate between the four phases of plant health decline using physiological variables and NiRA data. Both Gradient Boosting Trees (GBT) and Deep Learning (DL) showed the best performances, achieving favourable accuracies of 96.8% and 91.2% without NiRA data, 84.6% and 88.2% with NiRA data, with shorter training times. Using thermal technology weighted amongst the highest of the best performing variables using GBT, the utility and accuracy showed good discrimination between the stages of plant decline and is encouraged for future research in this field.
Abstract Understanding the allometric relationships between leaf area and other plant traits is essential for non-destructive growth monitoring and efficient crop management. However, no comprehensive study has yet modeled leaf area in quinoa ( Chenopodium quinoa Willd.) using simple morphological traits across different sowing dates. This study aimed to quantify allometric relationships between leaf area and plant height, leaf dry weight, stem dry weight, panicle dry weight, and total dry matter, and to evaluate whether these relationships are modified by sowing date. A two-year field experiment was conducted with 12 sowing dates under a randomized complete block design with three replications. Leaf area index (LAI) dynamics were described using a logistic model, and allometric relationships were fitted using power-law equations. The results showed that LAI followed a logistic trend across all sowing dates, with maximum values ranging from 2.7 to 7.9. Plant height provided the most reliable prediction of leaf area (R² = 0.83, b = 1.2), followed by leaf dry weight (R² = 0.72, b = 0.97). The allometric coefficients for stem dry weight (b = 1.54, R² = 0.74) and panicle dry weight (b = 1.95, R² = 0.71) showed greater variability. A striking finding was the exceptionally high allometric coefficient (b = 4.95) recorded on May 6 of the second year, indicating a pronounced shift in resource allocation toward leaf area expansion. Total dry matter was a weak predictor (R² = 0.54), likely due to leaf fall during the growing season. The hypothesis that sowing date modifies allometric relationships was confirmed, as evidenced by considerable variation in allometric coefficients across sowing dates. This study provides, for the first time, a comprehensive set of allometric models for quinoa across multiple sowing dates. Plant height and leaf dry weight are recommended as simple, rapid, and non-destructive indicators for leaf area estimation, facilitating improved crop monitoring and management under diverse environmental conditions.
Soil salinity is a major constraint on rice ( Oryza sativa L.) production, particularly during the yield-determining reproductive stage. Utilizing a high-throughput RGB platform, we non-destructively phenotyped a diverse panel of 294 rice accessions over two consecutive years. By extracting 60 dynamic image-based traits (i-traits) reflecting canopy architecture and stay-green capacity, and four seed-setting rate-related traits, our genome-wide association study (GWAS) identified 95 significant loci, 35.8% of which precisely co-localized with previously reported QTLs. We further prioritized OsSLT1 ( LOC_Os01g05790 ) as a candidate gene at a reproducible suggestive locus associated with leaf-rolling-related image traits. Transgenic evaluations confirmed that it acts as a positive regulator of salt tolerance at the seedling stage. OsSLT1 was mainly detected in the nucleus, and no significant changes in Na + or K + accumulation were observed in flag leaves under the tested salt-stress condition, suggesting that OsSLT1 may regulate salt tolerance through mechanisms beyond classical shoot ion accumulation. Natural variations in the OsSLT1 promoter were associated with transcriptional divergence. The Hap2 promoter haplotype showed significantly higher stress-induced transcriptional activity. Hap2 was rare in modern indica accessions, suggesting that it may represent a potentially useful genetic resource for future salt-tolerance improvement. Supplementary information The online version contains supplementary material available at 10.1007/s11032-026-01704-2.
Image-based plant disease identification is essential for advancing smart agriculture, particularly for staple crops such as rice and economically significant crops such as tea, which are cultivated under complex environmental conditions. Furthermore, these crop groups present distinct challenges: rice leaf disease data typically exhibits clear pathological structures but necessitates large-scale deployment on resource-limited devices, whereas tea leaf disease data is complicated by variable lighting, diverse backgrounds, and high biodiversity. Although current deep learning models achieve high accuracy, they predominantly utilize deep convolutional neural network (CNN) architectures with millions of parameters, which hinders practical deployment on edge devices and increases the risk of overfitting when field data is scarce. In response, this study introduces DisQuan, a hybrid architecture that integrates classical deep learning with quantum machine learning (QML) to balance accuracy and resource efficiency. In particular, DisQuan combines the lightweight DisNet feature-extraction network with a variablequantum neural network to compress and refine feature representations in quantum space, yielding a model with only 0.09 million parameters. Experimental results on rice and tea leaf disease datasets indicate that DisQuan achieves the highest accuracy on the rice dataset and performance comparable to deep CNN models with significantly more parameters on the tea dataset, while maintaining a compact and stable structure. Overall, these findings suggest that DisQuan provides a practical compromise between performance and model complexity, and highlight the potential of quantum-classical hybrid architectures for plant disease detection in real-world agricultural settings and on resource-constrained devices.
Manual inspection of grain plant leaves for defects is subjective and labor-intensive. Few studies have compared deep learning methods on a combined multi-crop dataset. The study collected locally 5,640 leaf images from rice, maize, and guinea corn farms in Nigeria and grouped them into six classes representing defective and healthy leaves for each crop. Three models were trained: YOLOv8 for end-to-end detection and classification, EfficientNetB0 for standalone image classification, and a hybrid that used YOLOv8 for leaf detection followed by EfficientNetB0 for patch classification. The hybrid achieved 99.85% accuracy on the test set, slightly above EfficientNetB0 (99.82%) and YOLOv8 (mAP 0.995). The hybrid also supplies bounding box locations, helping farmers identify exactly where damage appears. This system offers a reliable, field-deployable tool for monitoring grain crop health.
Field / plotLeafSegmentationDisease symptoms / severity
Camellia oleifera leaf disease segmentation under natural field conditions is important for precision plant protection but remains challenging because lesions often show small target areas, blurred boundaries, uneven illumination, complex backgrounds, and coexisting symptoms. To address these problems, this paper proposes BFMambaNet, a Boundary-Frequency-guided Global Semantic Mamba Network for fine-grained disease segmentation. The model adopts an encoder-decoder framework and introduces a Global Semantic Mamba-based spatial selective feature modeling block to capture long-range lesion context and reduce semantic confusion. A gated wavelet spatial enhancement block is further designed to strengthen high-frequency boundary details while suppressing noisy responses. During training, boundary-frequency auxiliary supervision guides contour localization and pathological texture recovery without additional manual boundary labels. A reinforcement-learning-guided adaptive loss controller adjusts class-wise reweighting factors and loss-component weights according to the training state, improving optimization stability. A pixel-level dataset containing 1400 images and seven disease categories was constructed for evaluation. Experimental results show that BFMambaNet achieves 92.39% Precision, 91.43% Recall, 91.26% Dice, and 85.46% mIoU, outperforming representative CNN-based, Transformer-based, and Mamba-based models. Evaluations on environmental subsets confirm superior robustness, outperforming VMamba by 3.70% mIoU under uneven illumination, 3.55% mIoU under complex backgrounds, and 5.10% mIoU under coexisting symptoms. Cross-dataset validation on Apple leaf diseases further proves its generalization with 3.39% mIoU and 3.84% Dice improvements over U-Mamba, while maintaining a competitive inference speed of 30 FPS. Qualitative results also show clearer boundaries, fewer missed small lesions, and more stable predictions in complex field scenarios.
Summary Understanding how plant populations respond to environmental variation through functional leaf traits remains challenging due to limitations of traditional phenotyping approaches. Hyperspectral reflectance offers a powerful high‐throughput solution, simultaneously capturing leaf biochemistry, water content, and structural properties across hundreds of wavelengths. We present a framework combining hyperspectral data, inverse modeling, and network analysis to investigate population‐level variation in Streptanthus tortuosus . Using a common garden experiment with four populations, we apply supervised methods (partial least square discriminant analysis; ridge regression) to identify which spectral features differ among populations, and an unsupervised spectral network approach to characterize how wavelength correlations are organizationally structured within each population, where we treat coordination architecture itself as a population‐level phenotype that can vary with environment. The framework detects distinct, heritable spectral signatures across populations, population differences in anthocyanins, carotenoids, Chl, water content, and population‐specific network architectures. Thermally variable environments were associated with greater spectral modularity, demonstrating that trait coordination architecture varies with climate of origin. This approach addresses the phenotyping bottleneck in evolutionary ecology, providing a scalable, high‐throughput tool for characterizing genetically based population differences in both individual traits and their coordination, with broad applications for monitoring plant population responses to climate change.
Accurate field detection of candidate tea shoots could support plantation monitoring, yield estimation, fresh-leaf assessment, and future selective-harvesting research. Wuyi rock-tea shoots are small and slender, have weak visual boundaries, and are easily confused with branches, petioles, and complex canopy backgrounds. Here, we developed YOLO11s-CSNG for candidate shoot detection in natural plantation scenes. The model combines a channel-spatial feature enhancement bottleneck, a normalized Wasserstein distance constraint for bounding-box regression, and ghost convolution layers in the detection head. We evaluated the model through detector comparisons, module ablations, and repeated training with five matched random seeds on a natural-scene dataset containing four Wuyi rock-tea cultivars. Across the five matched seeds, the mean mAP@0.5 increased from 67.37 ± 0.91% to 67.97 ± 0.90% on the validation set and from 61.29 ± 0.40% to 61.88 ± 0.66% on the internal test set. Neither paired difference was statistically significant: The 95% confidence intervals included zero, and the exact two-sided paired-permutation p values were 0.375 and 0.250, respectively. The mean mAP@0.5:0.95 did not improve. YOLO11s-CSNG retained a model size and model-only edge-inference time comparable to YOLO11s, providing a compact design for candidate shoot-region detection under the sampled field conditions.
Reproduction assets foundThe paper's phenotyping analysis is based on the publicly available PlantVillage plant leaf disease image dataset hosted on Kaggle (54,303 labeled leaf images across 38 classes), which the authors explicitly state was sourced from a publicly available Kaggle dataset. No author-specific code, models, or derived datasetsDataset · publicThe research incorporated PlantVillage dataset(24)
accessible on Kaggle that contains 54,303 plant leaf
images showing both healthy and diseased
conditions spanning across 38 specific categories.Open asset ↗Kagglepdf-raw-page:2 lines:1-105Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Abstract The classification of banana leaf disease has a large impact on agricultural output and relies heavily on timely early detection, with reliability as a fundamental component of effective crop management. The framework proposed in this study comprises a Hamiltonian Full Node Coverage Graph Attention Network (HFNC-GAT) and Fuzzy C-Means super pixel graph learning to explain the classification of banana leaf diseases. The HFNC-GAT allows for the representation of segmented leaf areas as the nodes in a graph. This framework also makes optimal use of an attention learning model to represent the spatial dependence of diseased leaf regions, allowing it to leverage both local and global spatial dependencies. The HFNC-GAT was demonstrated through observed experiments to achieve high performance with 96.11 and 94.19 accuracy, 0.9111 Cohen's Kappa, 0.9111 MCC, 0.9344 F2-score, and 0.9939 ROC-AUC compared to the performance of conventional CNN, GCN, and baseline GAT models.
Reproduction assets foundThe paper's Dataset Availability statement explicitly declares two public Kaggle banana leaf image datasets used for the phenotyping/classification experiments: Banana Leaf Disease Dataset V4 and BananaLSD. No author analysis code or trained model is reported as publicly available.Dataset · publicThe Banana Leaf Disease
Dataset V4 is available at https://www.kaggle.com/datasets/rayhanarlistya/banana-leaf-disease-dataset-v4.Open asset ↗Kaggle · banana-leaf-disease-dataset-v4pdf-page:19 lines:1-55Dataset · publicThe Banana Leaf Spot Diseases (BananaLSD) dataset is available at
https://www.kaggle.com/datasets/shifatearman/bananalsdOpen asset ↗Kaggle · bananalsdpdf-page:19 lines:1-55Code / dataset availability confirmedCrossref · checked 11 Sept 2026
Abstract Purpose Long-term monitoring of crop biophysical and biochemical traits remains challenging in high-latitude regions due to short growing seasons, frequent cloud cover, and highly variable weather. In this context, unmanned aerial vehicles (UAVs) offer flexible, high-resolution observations, but their added value relative to low-cost proximal sensors and their effectiveness for radiative transfer model (RTM) inversion across diverse crop canopies remain insufficiently quantified. This study evaluated the potential of a two-band proximal spectral reflectance sensor (SRS) and a five-band multispectral UAV sensor for retrieving leaf area index (LAI), leaf chlorophyll content (LCC), and canopy chlorophyll content (CCC) using PROSAIL inversion across major crops in Northern Europe over two growing seasons (2023–2024). Methods and Results Two inversion approaches – look-up table (LUT) and artificial neural network (ANN) were applied to PROSAIL simulations. UAV–PROSAIL–ANN outperformed LUT-based inversion and SRS observations, achieving the highest accuracy for LAI (R 2 = 0.81–0.95; RMSE = 0.27–0.77 m 2 /m 2 ), followed by CCC (R 2 = 0.58–0.94; RMSE 2 ), while LCC remained less accurately estimated (R 2 = 0.26–0.78; RMSE 2 ). Across sensors and methods, retrieval accuracy decreased in the order of LAI, CCC, and LCC, reflecting the stronger spectral control of canopy structure compared to biochemical traits. Conclusions The UAV–PROSAIL–ANN framework effectively captured spatial and temporal variability in crop traits, producing canopy-scale maps consistent with field observations. These results demonstrate the robustness and scalability of hybrid PROSAIL–ANN inversion for high-latitude crop monitoring, while highlighting current limitations in biochemical trait retrieval using multispectral data.
Reproduction assets foundThe paper's Data availability statement explicitly deposits the authors' UAV image processing code (irradiance normalization, vignetting, exposure compensation, radiometric calibration) in a public GitHub repository under GPL v3.0; other data are available only upon request.Code · publicData availability Code to perform irradiance normalization, vignetting, exposure compensation, and radio-
metric calibration is available at https://github.com/fieldSITES/scripts/tree/main/UAV under GNU General
Public License v3.0. Other data will be made available upon request.Open asset ↗UAVpdf-page:34 lines:1-40Plant phenotyping relevance matchCrossref · checked 11 Sept 2026
Md Nahidur Rahaman · Abdullah Al Mamun · Md. Kamal Hossen · Abdur Rouf · Tumpa Rani Shaha · Jungpil Shin · Mohd Nizam Husen · Abu Saleh Musa Miah
RiceLeafClassificationDisease symptoms / severity
Rice leaf diseases significantly affect crop health and yield potential, creating a need for accurate and timely disease diagnosis. Many existing approaches still rely on single-stream feature extraction architectures, which may limit the ability to simultaneously capture global contextual information and fine-grained disease characteristics. Moreover, limited interpretability and decision-support capability hinder their practical deployment in real-world rice farming. To address these limitations, we employed a framework consists of two parallel feature extraction streams designed to capture different characteristics of disease patterns. The first stream uses a Swin Transformer to learn global contextual information and long-range spatial relationships across the leaf image. The second stream employs ConvNeXt to extract local texture features, including lesion details, spots, and color variations. By combining these complementary representations, the proposed framework effectively integrates global semantic information with local disease-specific features for improved classification performance. The extracted features from the two streams are fused through concatenation followed by an attention-based feature refinement module, enabling adaptive weighting of discriminative features. The refined representation is then used by a fully connected classifier for disease prediction. To enhance model interpretability, Grad-CAM visualization is incorporated to highlight disease-relevant regions and provide visual explanations for the model decisions. Furthermore, an LLM-based advisory module is integrated as a post-diagnosis decision-support component to provide contextualized disease management information and suggestions based on the predicted disease category. The generated suggestions are intended to support, rather than replace, expert agronomic recommendations and should be validated by agricultural professionals before practical application. The proposed framework was evaluated on two rice leaf disease datasets, achieving accuracies of 99.55% and 97.06% on Dataset-1 and Dataset-2, respectively, which are higher than those reported in previous studies. Additionally, five-fold cross-validation on Dataset-1 achieved an average accuracy of 98.91% ± 0.47, demonstrating the stability of the proposed approach. Cross-dataset evaluation using nine common disease classes across both datasets achieved 90.80% accuracy, indicating improved generalization across different data distributions. The proposed framework provides an accurate and explainable approach for rice leaf disease diagnosis in smart agriculture applications.
This paper explores how deep learning methods can be used to monitor the health of crops and identify diseases, particularly for the apple crop. As the need for food security and sustainable farming methods increases, there is a strong demand for early detection of crop diseases. We have used a Convolutional Neural Network (CNN), based on the model of VGG16 architecture, since the model is known to be effective in image classification. The dataset contained 7771 training images for to enhance machines deep learning regarding plant diseases. Following that, validation image collection of 1747 images divided into four health conditions of the apple crops. In addition, the model was evaluated using 196 new images as a final test. To enhance the model capacity for recognizing diseases of real leaves rather than just memorize exact training pictures, data augmentation was used with ImageDataGenerator of TensorFlow. This means the training images were zoomed, rotated, and shifted to enable the machine detects more variations. Ten epochs of training were performed to measure the model accuracy. The results indicated that the model obtained significant improvement in training and validation accuracy from 56.43% to 78.12% and 92.94 to 96.93%, respectively. Most impressively, the final test dataset, which contained completely new images, scored an accuracy rate of 98%. The results indicate that in the architecture field the application of deep learning methodologies is effective, suggesting that automated detection of diseases by using sophisticated image analysis manages crop diseases identification efficiently. Combination of these methods successfully creates avenues for novel research to built real-time systems of crop disease monitoring, which help farmers increase their productions and farm their lands sustainably.
High-throughput phenotyping is key in modern breeding for rapidly and cost-effectively evaluating salt-adaptive traits. However, few studies have combined spectral reflectance indices (SRIs) with deep learning to assess field-grown wheat under salt stress. In this study, we developed optimized 2D and 3D SRIs integrated with artificial neural network (ANN) to assess chlorophyll a (Chl a), chlorophyll b (Chl b), total chlorophyll (TChl), and grain yield (GY) in 32 recombinant inbred lines (RILs) and four cultivars under 150 mM NaCl field conditions. ANOVA revealed that genotype contributed the largest proportion of the treatment sum of squares across all traits (60–80%), followed by the genotype × year interaction (8–15%), whereas year contributed the smallest proportion (1–5%). Heatmap clustering of these traits clearly distinguished salt-tolerant from salt-sensitive genotypes. The study findings highlight using four key traits as screening criteria for salt tolerance in wheat. Our optimized 2D/3D spectral indices showed moderate to strong predictive power (R 2 = 0.25–0.75), outperforming earlier indices. Multi-season data improved accuracy by 15–25%, with best predictions for Chl a and TChl (R 2 = 0.34–0.75) versus Chl b and GY (R 2 = 0.25–0.64). Top models included ANN-3D-SRIs-8 for Chl a (R 2 = 0.735/0.644), ANN-3D-SRIs-3 for Chl b (R 2 = 0.611/0.549), ANN-2D-3D-SRIs-3 for TChl (R 2 = 0.713/0.619), and ANN-2D-SRIs-2 for GY (R 2 = 0.648/0.553). This framework combines optimized indices and machine learning for scalable, high-throughput phenotyping to advance precision breeding of salt-tolerant wheat.
Nitrogen is an important element present in vital substances such as plant proteins and chlorophyll, the content of which directly reflects the nutrient status of crops, and provides a theoretical basis for crop nutrient diagnosis, growth monitoring and yield potential prediction. Taking the chip-level visible/near-infrared spectral sensor AS7263 as the data acquisition module and the Arduino Uno single-chip microcomputer development board as the control module, a portable crop leaf spectral sensing system was designed in this study. The spectral reflectance and nitrogen content of wheat leaves were obtained through field experiments. Principal Component Analysis (PCA) was used to eliminate abnormal spectral data. Combined with pretreatment algorithms including Multiplicative Scatter Correction (MSC) and Standard Normal Variate (SNV), the prediction models for wheat leaf nitrogen content were established based on Partial Least Squares (PLS), Support Vector Machine (SVR), Random Forest (RF) and a Back Propagation (BP) neural network. The results showed that compared with SNV, the model performance based on the spectral data after MSC pretreatment was better. The test set R 2 values of PLS, SVR, RF and BP models were 0.61, 0.75, 0.83, and 0.89, and the root mean square errors (RMSEs) were 4.62 mg g -1 , 4.38 mg g -1 , 3.39 mg g -1 and 3.27 mg g -1 , respectively. The MSC-BP prediction performance was the best, and the non-destructive and accurate detection of nitrogen in wheat leaves was realized, which verified the feasibility of micro-spectral sensing technology in crop nutrition diagnosis.
O cultivo de maçãs tem grande importância econômica no setor agropecuário brasileiro, especialmente na região Sul do país. No entanto, a produtividade dos pomares é frequentemente comprometida por doenças foliares que, se não tratadas, podem resultar em perdas substanciais. Nesse contexto, os avanços em técnicas de Aprendizado de Máquina têm possibilitado o desenvolvimento de soluções computacionais que auxiliam no diagnóstico agropecuário com maior precisão e agilidade. Este trabalho propõe uma abordagem baseada em Redes Neurais Convolucionais que utiliza aprendizado por transferência para detectar automaticamente sintomas de doenças em folhas de macieira. A metodologia desenvolvida inclui a segmentação e análise de regiões sintomáticas para reduzir o ruído proveniente de áreas saudáveis e direcionar o aprendizado do modelo para sinais relevantes. Um total de 32.382 manchas de sintomas foram extraídas de 1.995 imagens originais, abrangendo cinco classes de distúrbios foliares: glomerela, sarna, danos por herbicidas, deficiência de magnésio e deficiência de potássio. A rede MobileNetV2, treinada por meio de aprendizado por transferência, alcançou um F1-score de 0,926 e 93,8% de acurácia no conjunto de teste reservado. Os resultados indicam um bom desempenho no contexto avaliado, sugerindo o potencial da abordagem como ferramenta de apoio ao diagnóstico da saúde das plantas e à tomada de decisões em campo. Palavras-chave: aprendizado de máquina; visão computacional; doenças em plantas. Application of convolutional neural networks for disease detection in apple tree leaves ABSTRACT: Apple cultivation holds significant economic importance in the Brazilian agricultural sector, especially in the southern region of the country. However, orchard productivity is often compromised by foliar diseases, which, if left untreated, can lead to substantial losses. In this context, advances in Machine Learning techniques have enabled the development of computational solutions that support agricultural diagnostics with greater accuracy and agility. This work proposes an approach based on Convolutional Neural Networks that uses transfer learning to automatically detect disease symptoms in apple leaves. The developed methodology includes segmentation and analysis of symptomatic regions to reduce noise from healthy areas and focus the model’s learning on relevant signals. A total of 32,382 symptom patches were extracted from 1,995 original images, covering five foliar disorder classes: glomerella, scab, herbicide damage, magnesium deficiency, and potassium deficiency. The MobileNetV2, trained via transfer learning, achieved a F1-score of 0.926 and 93.8% accuracy on the held-out test set. The results indicate good performance in the evaluated setting, suggesting the approach’s potential as a tool to support plant-health diagnosis and field decision-making. Keywords: machine learning; computer vision; plant disease.
Understanding the dynamic regulation of endogenous metabolites in plants under environmental stress is essential for elucidating plant adaptive mechanisms and improving crop resilience. However, conventional analytical methods are typically destructive, time-consuming, and lack the capability for real-time monitoring, thereby limiting the investigation of in-vivo biochemical dynamics in plants. In particular, the in-situ, non-invasive detection of small-molecule regulators such as protocatechuic acid (PCA) remains a significant challenge. Herein, we report a wearable electrochemical sensing platform based on a Cu/ZIF-8-modified flexible printed electrode (FPE) for real-time, in-situ monitoring of PCA in plant leaves. The incorporation of Cu into the ZIF-8 framework enhances the electrical conductivity and electrocatalytic activity of the material while maintaining its porous structure, enabling sensitive detection of PCA. By integrating with reverse iontophoresis (RI), non-invasive extraction and continuous monitoring of PCA from living plant tissues are achieved. The dynamic behavior of PCA in green tea plants under light deprivation and drought stress is systematically investigated. The results reveal distinct stress-dependent response patterns, with PCA levels rapidly decreasing under both dark and drought conditions, highlighting its critical role in stress adaptation and metabolic regulation. This work establishes a versatile strategy for real-time tracking of endogenous plant metabolites and provides new insights into plant physiological responses under environmental stress. The proposed platform holds significant promise for applications in plant science, precision agriculture, and the development of stress-resilient crops.
Automated plant identification based on leaf morphology has gained significant attention in recent years due to its wide range of applications in precision agriculture, biodiversity conservation, environmental monitoring, and botanical informatics. Advances in digital image processing and machine learning have enabled the development of intelligent systems capable of identifying plant species from leaf characteristics with minimal human intervention. Despite these advancements, achieving reliable and accurate classification remains challenging because leaf images are often affected by variations in illumination, complex backgrounds, image noise, differences in orientation and scale, as well as natural leaf deformation. These factors can obscure important morphological features, reduce the effectiveness of feature extraction, and ultimately decrease the accuracy and robustness of automated plant classification systems. Consequently, there is a growing need for intelligent frameworks that can effectively handle these challenges while preserving critical leaf morphology and venation information for reliable plant identification. This study proposes an Intelligent Morphology-Driven Framework that integrates advanced digital image processing and machine learning for robust leaf venation analysis and plant classification. The proposed framework integrates multiple digital image processing and machine learning techniques to enable accurate and automated leaf venation analysis and plant classification. Initially, leaf images undergo preprocessing using grayscale conversion, histogram equalization, Contrast Limited Adaptive Histogram Equalization (CLAHE), Gaussian filtering, Laplacian sharpening, Gabor filtering, and homomorphic filtering to improve image quality and enhance venation and structural details. The enhanced images are then processed through threshold-based segmentation followed by morphological operations, including erosion, dilation, opening, closing, convex hull generation, and skeletonization, to accurately isolate leaf regions while preserving their geometric structure.To characterize leaf morphology, the framework extracts a comprehensive set of features, including geometric descriptors such as area, perimeter, circularity, aspect ratio, solidity, eccentricity, and vein density, together with Hu invariant moments that provide rotation-, translation-, and scale-invariant shape representation. In addition, the framework investigates the influence of image compression by comparing lossless PNG and lossy JPEG formats to evaluate their impact on preserving morphological features and venation details. The extracted feature vectors are subsequently classified using a Random Forest classifier to categorize leaf venation patterns into parallel, reticulate-pinnate, and reticulate-palmate classes.Experimental evaluation demonstrates that the proposed framework achieves an overall classification accuracy of 93.2%, while effectively preserving important morphological characteristics and maintaining computational efficiency. The combination of adaptive image enhancement, morphology-preserving segmentation, comprehensive feature extraction, and robust machine learning classification makes the proposed approach reliable, interpretable, and scalable. Consequently, the framework has significant potential for applications in digital herbarium systems, automated plant identification, biodiversity monitoring, botanical informatics, and precision agriculture.
Abstract Sustainable agriculture has substantial share on improvement of food security and optimization of resources utilization particularly for high value crops like rice leaf. Rice varieties should be properly classified in order to benefit the harvest management, reduced loss after harvest and improved agriculture methods. The traditional classification method usually brings the low precision and the traditional classification method is also subjected to human error, which is difficult to bring about reliable output. This study introduces an optimized multi-class rice leaf disease classification system utilizing “Rice Feature Selection” (RiceFS) and ensemble machine learning approaches. RiceFS is realized based on a feature selection mechanism based on Recursive Feature Elimination. Selected classifiers such as KNN, Random Forest, Gradient Boosting, Ensemble Learning and Optimized SVM are analyzed based on the extracted subset of features and the proposed system is used to classify the seven classes of rice leaf disease. The experimental results show that the Optimized SVM has the best classification results among the different classifiers with accuracy of 92.10%, Precision of 92.20%, balanced Recall and F1 Score, which shows that Optimized SVM is very effective in multi-class rice leaf disease classification. The performance can be improved by feature reduction, generalization capability and computational complexity reduction, which are realized with the help of RiceFS. The proposed framework is designed to provide an intelligent decision support system for timely intervention, loss minimization and sustainable agriculture. Results indicate that these algorithms are applicable for rice leaf disease classification since they are accurate, reliable and scalable.
Reproduction assets foundThe paper's RiceFS phenotyping/classification experiments are built on two public Kaggle rice leaf disease image datasets, explicitly cited with URLs and a data availability statement. No author code or models are deposited.Dataset · publicThe RiceFS framework proposed initially
performs a feature selection, followed by training
several classifiers: K-Nearest Neighbors (KNN),
Random Forest (RF), Gradient Boosting (GB),
Ensemble Learning, and Optimized Support
Vector Machine (Optimized SVM). The data is
published on the Kaggle website. The data is
open-source at:
https://www.kaggle.com/datasets/vbookshelf/rice-leaf-diseases [44].
This data consists of 120 jpgs of disease infected
rice leaves. The photos are divided into 3
categories according to the kind of disease. There
are 40 images in each class.
Classes
• Leaf smut
• Brown spot
• Bacterial leaf blight
The datasets are preprocessed by eliminating
redundant information, normOpen asset ↗Kaggle · vbookshelf/rice-leaf-diseasespdf-raw-page:11 lines:1-103Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
India’s economy is heavily reliant on agriculture, with a diverse array of crops grown on vast tracts of land. Fruit cultivation, especially papaya, has become more popular in recent years because of its high nutritional and financial value. To increase yield, maximize resource use and lessen reliance on chemical pesticides, modern techniques like protected cultivation and hydroponics are being used more and more. Fruit crops grown in controlled or semi-controlled environments are still susceptible to nutrient imbalances despite these developments, which can have a substantial impact on plant health, fruit quality and overall productivity. Papaya leaf nutrient deficiencies frequently show up in the early stages of growth and can result in poor fruit development and decreased yield if they are not detected in time. To support early diagnosis and better crop management, the current study focuses on creating an effective method for identifying nutrient deficiencies in papaya leaves using a deep learning (DL) framework based on transfer learning (TL). In this nutrient and micronutrient deficiency study and field work observation during year 2024 to 2026 with different climate and weather conditions done in order to tackle a new but related classification task, in the context of plant health assessment, several well-established architectures including InceptionV3, VGG19, DenseNet and Xception have been widely explored for leaf image analysis. Studies commonly utilize publicly available datasets, such as papaya leaf image repositories hosted on platforms like IEEE DataPort, to fine-tune these models for efficient feature extraction and accurate identification of nutrient and micronutrient deficiency patterns. This body of work demonstrates the growing role of transferring convolutional neural network (CNN) models in advancing automated crop monitoring and decision support systems.
Accurate estimation of the photorespiratory CO 2 compensation point (Γ*) is essential for describing the balance between Rubisco carboxylation and oxygenation and for parameterising biochemical models of photosynthesis. Γ* and the rate of CO 2 release in the light (D L ) are commonly estimated using the Laisk method, based on measurements of net CO 2 assimilation rate (A net ) at low chloroplastic CO 2 concentrations (c c ), under several sub-saturating irradiance levels. However, many widely used temperature dependence relationships for Γ* (Γ*(T)) were derived using conventional linear implementations of the Laisk method, despite the intrinsically nonlinear behaviour of the A net -c c response predicted by the photosynthetic theory. Here, we revisited the temperature dependence of Γ* and D L using the improved Laisk-FvCB framework that simultaneously constrains the nonlinear A net -c c response across multiple irradiance levels. Gas exchange of sunflower leaves was measured across a wide temperature range from 3.9°C to 42.0°C. The conventional linear implementation generated highly dispersed pairwise intersections and unstable estimates of both Γ* and D L , including some physiologically unrealistic negative D L values at low temperatures. In contrast, the mechanistically constrained Laisk-FvCB framework produced physiologically meaningful temperature responses and substantially reduced methodological artefacts associated with linear extrapolation. Using this framework, we derived a revised in vivo Γ*(T) relationship described by an Arrhenius-type function with Γ*(25) = 43.4 μmol mol -1 and an apparent activation energy of 27.7 kJ mol -1 , such that Γ*(T) = 43.4 exp[11.176 ((T - 25)/(T + 273.15))], where T is leaf temperature in °C. Comparison with other widely used Γ*(T) formulations showed substantial divergence at temperature extremes, often exceeding the variability expected from realistic interspecific differences in Rubisco specificity among C 3 species.
Corn is a staple crop of global significance; however, foliar diseases may lead to 30–60% yield loss if not detected at an early stage. Conventional visual inspection is time-consuming, subjective, and difficult to scale for smallholder farmers worldwide. To overcome these issues, we present Corn Transfer Learning Network (CTL-Net), an end-to-end hybrid deep learning model for corn leaf disease identification. CTL-Net, which combines Inception-ResNet-v2 as the backbone and MobileNetV3 as a feature extractor in parallel convolutional streams, simultaneously learns diverse scales of texture information from low-level textures, mid-level structural patterns, and high-level disease semantics of RGB leaf images. Adaptive feature fusion is formulated through learnable weighting coefficients and bi-directional spatial–channel attention mechanisms, which further enhance feature discriminability and robustness. The proposed approach is tested on an extensive dataset of 12,456 images from 10 corn diseases, including Northern Leaf Blight, Common Rust, Gray Leaf Spot, and Cercospora Leaf Spot, acquired under controlled and real-field conditions. CTL-Net attains the highest classification accuracy of 99.42%, outperforming DenseNet121 (97.92%), EfficientNetB3 (97.35%), and general stacking models (97.89%). Robustness experiments demonstrate the effectiveness of the proposed method against illumination variations, additive noise, and partial occlusions. CTL-Net enables real-time inference with a latency of 42 ms on an NVIDIA RTX 3090 GPU. Gradient-weighted Class Activation Mapping++ (Grad-CAM++)-based interpretability analysis results in a mean Intersection over Union (IoU) of 87.6% with expert-annotated disease regions. Five-fold cross-validation, ablation studies, and statistical significance testing (p
Reproduction assets foundThe paper states its final curated corn leaf disease dataset (12,456 images, 10 classes) is publicly available via the authors' GitHub repository vishruthkp/maizedataset (reference [30] and Data Availability statement). The Kaggle PlantVillage and Corn or Maize Leaf Disease datasets are cited source inputs, not paper-Dataset · public"Prediction of Crop Yield using Machine Learning," International Available: https://github.com/vishruthkp/maizedataset.Open asset ↗vishruthkp/maizedatasetpdf-page:7 lines:1-53Code / dataset availability confirmedEurope PMC · checked 5 Sept 2026
El Karch H, Natij Y, Benaly M, El Gouri R, Mezouari A.
Field / plotLeafClassificationStress / disease detectionDisease symptoms / severity
Automated plant disease recognition from field photographs remains challenging: models trained on laboratory datasets collapse on in-the-wild images, and the current state of the art on PlantWild-the largest open in-the-wild benchmark (18,542 images, 89 classes)-relies on text prototypes derived from a language model to reach 76.18% top 1 accuracy. We ask whether text supervision is truly necessary or whether the bottleneck is the diversity of the visual representation. Our central methodological finding is that backbone selection and fusion matter far more than classifier-head engineering : across 118 experiments, a simple fixed-weight linear-prototype combination on top of three complementary frozen backbones yields larger and more reproducible gains than any head-level adaptive routing mechanism we test. Specifically, through a systematic study of six frozen vision backbones (three CLIP, two DINOv3, and one DINOv2), three classifier heads, and four fusion configurations, completed in a single day on one consumer GPU, we establish three findings. (i) A single self-supervised backbone (DINOv2 ViT-L/14) already surpasses text-augmented MVPDR (77.56% vs. 76.18%). (ii) Concatenating three complementary backbones (DINOv2 + DINOv3 + CLIP) reaches 80.23% ± 0.41% (five seeds), exceeding the published MVPDR accuracy by +4.05 points and our own reproduction of MVPDR under an identical evaluation protocol by +7.96 points, without any language supervision. (The difference between the two deltas reflects evaluation-protocol differences-our split, model-selection criterion, and training schedule-rather than any discrepancy in the reported numbers; see Section 4.6.6 for a full reconciliation.) (iii) Every form of learned routing we test-per-class gating, backbone gating, sample-wise gating-is inessential; the gain is entirely attributable to backbone diversity and a simple linear-prototype scoring combination. On the smaller PlantDoc benchmark, the same principle transfers but with substantially higher seed variance: the best configuration reaches 80.09% at a favourable seed but 76.97% ± 1.48% over five seeds-a suggestive rather than robust gain. Beyond the accuracy headline, we provide a pathology-aware per-class analysis showing that DINOv2/v3 dominate on fine-texture lesion classes (rusts, mildews, and leaf spots) whilst CLIP's narrow advantage concentrates on organ/species-level identification (rice leaf and potato late blight). All primary claims are validated over five seeds, and all code, feature caches, and result files are released for full reproducibility.
Reproduction assets foundThe paper states that its code, cached features, and result files will be released, but no authors' public URL or repository is provided in the supplied blocks (future-tense availability language only). The DINOv3 GitHub link and OpenReview link are third-party backbone resources, not paper-specific assets.Code · publicAll primary claims are validated over five seeds, and all code, feature caches, and result files are released for full reproducibility.Open asset ↗lines:332-335Code / dataset availability confirmedCrossref · checked 14 Sept 2026
Published11 Aug 2026Engineering, Technology & Applied Science ResearchCited by 0 · OpenAlex ↗
In recent years, areca nut plants have been vulnerable to different diseases that appear as distinct colors on leaves, caused by bacteria or fungi. These symptoms disrupt photosynthesis and reduce yield, affecting productivity and crop health. Therefore, accurate plant disease classification is essential for detecting distinct disease shapes and sizes. Existing Deep Learning (DL) models have several limitations that prevent them from distinguishing between various plant diseases due to similar characteristics. To overcome this limitation, a Dynamic elastic boundary strategy Sailfish Optimization Algorithm and Convolution Neural Network (DSFO-CNN) method is proposed to identify and accurately classify arecanut plant diseases. The Visual Geometry Graph-19 (VGG-19) model extracts features that have significant information about disease in arecanut plants. The proposed arecanut plant disease classification model employed feature selection and drop cyclic learning rate, which adjusts the CNN learning rate to efficiently learn the subtle information about various leaf and nut diseases to enhance classification. The experimental results of the DSFO-CNN demonstrate superior performance compared to existing approaches.
Reproduction assets foundThe paper's phenotyping inputs are two public image datasets: the collected Arecanut dataset (Kaggle) and the PlantVillage dataset (Kaggle), both explicitly cited and declared openly available. No author analysis code or trained model is released.Dataset · publicDATA AVAILABILITY
The data used in this study are openly available at [19] and
[20].Open asset ↗pdf-page:7 lines:1-63Plant phenotyping relevance matchCrossref · Europe PMC · checked 15 Sept 2026
Accurate crop disease diagnosis is critical for agricultural productivity and food security, yet existing deep learning systems often struggle to generalize across visually similar diseases and varying environmental conditions. Recent Vision-Language Models (VLMs) have demonstrated promising zero-shot reasoning capabilities; however, most agricultural diagnostic systems still rely on isolated single-model predictions without collaborative reasoning or consensus mechanisms. In this work, we propose VIDA+PANDA, a multi-agent Vision-Language framework for zero-shot crop disease diagnosis. The framework consists of two stages: VIDA, where multiple VLM agents independently analyze crop leaf images to establish baseline performance, and the Peer-Anchored Named Deliberation Architecture (PANDA), which introduces a structured multi-round debate among a selected group of high-performing and architecturally diverse agents. During deliberation, agents exchange reasoning, critique peer predictions, and revise decisions through evidence-grounded discussion, while an anti-sycophancy mechanism discourages unsupported consensus shifts. Final predictions are generated through performance-weighted consensus voting. Experiments are conducted on the CDDM benchmark using seven heterogeneous VLMs from four independent providers, including two open-source models, under a fully zero-shot setting. A non-participant GPT-5 model serves as an independent judge to assess the final diagnostic predictions. Beyond conventional accuracy, the framework introduces three semantic measures: Semantic Label Similarity (SLS), which measures how semantically close a predicted crop-disease pair is to the ground truth and captures partial correctness overlooked by exact-match evaluation; Reasoning Specificity (RS), which measures how concretely an agent’s explanation references visual evidence such as lesion color, shape, texture, or margins; and Inter-Agent Reasoning Convergence (IRC), which measures the extent to which agents rely on similar visual evidence, capturing epistemic alignment independently of label correctness. Experimental results show that collaborative multiagent deliberation improves individual diagnostic performance and semantic alignment, with the largest gains observed among weaker participating agents. The findings also reveal important relationships between predictive accuracy, persuasive influence, and consensus formation in VLM-based agricultural diagnosis.
Raquel Shany Betancur-Choquehuayta · Esther Emiliana Molina-Gonzales · Dony Javier Flores-Subeleta · Raymundo O. Gutiérrez-Rosales · Alberto Anculle-Arenas · Natty Wilma Llasaca-Calizaya · José Luis Bustamante-Muñoz · Mayela Elizabeth Mayta-Anco
Within the context of climate change, quinoa ( Chenopodium quinoa Willd.) is a climate-resilient crop with high nutritional value. The effects of deficit irrigation on quinoa growth and physiological performance under arid conditions remain insufficiently understood. This study evaluated ten quinoa genotypes (two commercial varieties and eight accessions) under two irrigation regimes to identify traits and spectral indices associated with water-stress tolerance. We combined manual phenotyping of agromorphological and physiological traits with multispectral and spectroradiometer measurements to calculate 35 vegetation indices across 13 and 5 dates, respectively. Deficit irrigation reduced plant height (18%), specific leaf area (8%), yield (43%), harvest index (26%), relative water content (7%), and dry matter accumulation (36%), while relative chlorophyll content (SPAD, Soil Plant Analysis Development) and stomatal density increased by 16% and 13%, respectively; accession ACC_23 exhibited the highest water-use efficiency (5.9 g kg -1 ). A univariate analysis of 35 vegetation indices across 13 dates showed that: Health Index(HIV), Normalized Green-Red Difference Index (NGRD), Red-Green Ratio (RG) and Plant Senescence Reflectance Index (PSRI), were the most sensitive, detecting significant differences between irrigation treatments in up to 32 of the 130 possible genotype-by-date comparisons. Integrating remote sensing into crop phenotyping represented a significant methodological improvement by enhancing phenotyping efficiency, improving detection of deficit irrigation effects, and facilitating identification of tolerant quinoa genotypes for arid production systems.
Common beanLeafClassificationDisease symptoms / severity
Abstract Accurate recognition of plant leaf diseases from images is essential for intelligent agriculture and precision crop protection. However, reliable disease identification remains challenging because lesion regions often exhibit subtle visual differences, complex backgrounds, and large intraclass variations, especially when available disease samples are limited. To address these challenges, this study proposes PR-CNN, a deep learning framework that integrates convolutional neural networks, pyramid split attention, and a relation network for bean leaf disease image recognition. The convolutional backbone is first used to extract visual features from support and query images. Then, the pyramid split attention module enhances multiscale spatial and channel feature representation, enabling the model to focus on discriminative lesion regions while suppressing redundant background information. Finally, the relation network learns a nonlinear similarity metric between paired samples and generates relation scores for disease category prediction. Experimental results show that PR-CNN achieves an overall classification accuracy of 99.24% on the primary bean leaf disease dataset, outperforming representative models, including ResNet50, DenseNet, Inception v4, and EfficientNet B7, in terms of recognition accuracy and adaptability. In addition, PR-CNN was evaluated on four publicly available plant disease datasets, including CGIAR, Plant Diseases, LWDCD 2020, and Plant Pathology, achieving an average accuracy of 99.84%. These results demonstrate that PR-CNN can effectively improve image based plant disease recognition and provides a robust visual classification framework for intelligent crop disease diagnosis.
leaf development was tracked dynamically from high-throughput phenotyping image series using the SAM-2 deep learning model, enabling automated segmentation and tracking of individual leaves across hundreds of plants from different accessions grown under different water and nitrogen conditions. From these time series, leaf-level growth dynamics were reconstructed and used to calculate key developmental traits, including phyllochron and relative expansion rates. These measurements were further integrated into the GreenLab model to infer plant-scale parameters such as radiation use efficiency (RUE) and leaf demand parameters. Statistical analyses revealed significant genotype, treatment, and interaction effects on several traits, with stable genotype rankings but contrasted sensitivities to environmental conditions. While whole-rosette growth captured strong and consistent responses, individual leaf contributions to global growth remained limited, highlighting the integrative nature of rosette-scale dynamics. Model-derived parameters provided additional insight into plant strategies, showing that genotypes differ in both the duration and timing of leaf demand, with generally stable patterns across environments. In parallel, QTL mapping was performed using leaf-level measurements and phyllochron traits. This analysis identified shared and trait-specific genetic loci, including loci associated with leaf emergence dynamics that were not detected using traditional rosette-scale integrative traits. Overall, this study demonstrates that combining deep learning-based image analysis with mechanistic modeling enables large-scale and quantitative characterization of plant development, providing biologically interpretable traits and new insights into their genetic and environmental determinism. It highlights the potential of combining high-throughput image-based phenotyping, deep learning, and mechanistic modeling to generate biologically interpretable traits and to assess genetic and environmental influences on these traits.
Reproduction assets foundThe paper explicitly states that the authors' leaf segmentation software (AraLeaf_segmentation), the GreenLab Arabidopsis model (AraGreenlab), and the statistical analysis code (AraParameters_statistical_analyses) are open-source and publicly hosted on forge.inrae.fr, and that the data and results used in the paper areCode · publicThe leaf segmentation software (AraLeaf_segmentation) and the GreenLab model of Arabidopsis thaliana (AraGreenlab) are open-source software, and distributed under the GNU GPL v3 licence. The two software packages, and the code to run statistical analyses (AraParameters_statistical_analyses) are available at:https://forge.inrae.fr/phenoscope-public/plant_phenomics_suppmat.Open asset ↗phenoscope-public/plant_phenomics_suppmathtml-lines:419-444Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Citrus Huanglongbing (HLB) is one of the most destructive citrus diseases worldwide, and early diagnosis remains challenging because uneven pathogen distribution often leads to false-negative PCR results. Here, a paper-based nanozyme sensor array was developed for volatile organic compound (VOC) chemical fingerprinting of HLB-associated citrus samples. Fe/Al bimetallic NH₂-MIL-53 nanozymes were designed to regulate VOC adsorption and peroxidase-like catalytic activity. Exposure of the nanozyme sensors to VOCs reduced the catalytic oxidation of 3,3',5,5'-tetramethylbenzidine (TMB), generating concentration-dependent colorimetric responses. By tuning the Fe/Al ratio, the nanozymes exhibited differentiated responses toward HLB-associated volatiles, including methyl salicylate, phenylacetaldehyde, and linalool, with recognition limits of 0.1-0.5 ppm. Integration with two MOF-based sensing channels formed a five-channel artificial olfactory array capable of generating multidimensional color fingerprints. The sensor array successfully discriminated healthy, asymptomatic, and infected citrus leaf samples, achieving an overall classification accuracy of 92.5%. These results suggest that the proposed platform provides a simple and low-cost approach for citrus VOC chemical fingerprinting and may provide useful information for quality-related screening in citrus production systems, while further field-oriented validation is still needed to assess its practical applicability.
Fuente D, Zapata R, Martínez-Heras E, Raigón MD, Oliver-Villanueva JV.
Field / plotLeafStem / branchPhysiological trait estimationGrowth / time-series analysisStress response / toleranceWater status / transpiration
Live fuel moisture content is a key determinant of live fuel flammability, yet its destructive and discontinuous measurement limits high-temporal-resolution monitoring. This study evaluated whether leaf electrical potential can serve as a non-invasive proxy for LFMC and flammability-related traits under natural drought conditions. From February to July 2025, leaf and trunk electrical potentials were monitored weekly in Salvia rosmarinus individuals from a Mediterranean shrubland, while LFMC, essential oil yield, fatty-acid fraction, and laboratory-based flammability metrics-ignition time, combustion duration, and flame height-were assessed bi-weekly. Leaf electrical potential was strongly associated with LFMC (R 2 = 0.64, p < 0.001), decreasing as plants underwent seasonal drought-induced dehydration. Periods of high temperature and low rainfall reduced both LFMC and electrical potential, coinciding with shorter ignition times, which declined to approximately 20-30 s during the driest period. Based on the observed shifts in ignition time, combustion duration, and flame height, three empirical LFMC response zones were identified, with leaf electrical potential closely tracking transitions in plant hydration and flammability. These results suggest that plant electrophysiology may provide a promising non-invasive indicator of live fuel water status and seasonal flammability dynamics, with potential applications in wildfire risk monitoring when combined with conventional LFMC, meteorological, and remote-sensing approaches.
Background: Plant diseases significantly threaten global food security, reducing potential harvests and sometimes causing total crop failure. Traditional detection methods, which rely on manual inspection and laboratory testing, are time-consuming, costly and prone to human error. Methods: To address these challenges, this study applies transfer learning techniques using deep convolutional neural networks for accurate and efficient plant disease detection. A comparative analysis of nine pretrained models, VGG16, VGG19, ResNet50, ResNet101V2, MobileNetV2, InceptionV3, DenseNet121, InceptionResNetV2 and Xception was conducted on the PlantVillage dataset, focusing on apple, potato and peach leaf images. Result: Results show that DenseNet121 and ResNet101V2 achieved the highest accuracy, particularly for potato leaves with 98.5%, while MobileNetV2 also performed well with up to 99% accuracy for apple and peach leaves. The study demonstrates that transfer learning effectively enhances plant disease classification, enabling faster, more reliable and resource efficient detection for precision agriculture.
The tomato plant is considered one of the most important crops in the world, yet it is vulnerable to various diseases that affect crop quality and agricultural productivity. These challenges have driven the need for an efficient and intelligent plant disease detection system. With the development of computer vision and artificial intelligence, this proposed methodology based on deep learning for tomato leaf diseases has been presented. Two public datasets: Taiwan DS with nine classes and Tomato Leaf Diseases Detection Computer Vision Dataset (TLDDCV DS) with seven classes have been used to test this system. This system begins with plant image processing, which includes gamma correction and bilateral filtering, to enhance image quality and clarity while preserving key disease features. Then, a genetic metaheuristic algorithm was used to automatically select the most significant hyperparameters, further optimizing both processing time and accuracy. After that, the tomato leaf disease detection applies the You Only Look Once version 11 Nano (YOLOv11n) model. The YOLOv11n backbone is edited through a Data-efficient Image Transformer (DeiT) to improve the system's capacity for learning global contextual information and long-range dependencies. Experimental results demonstrate that the proposed system outperforms existing methods. It achieved an average mAP@50 of 97.8%, mAP@50-95 of 93.4%, precision of 97.3%, recall of 93.8%, and F1-score of 95.5% on the Taiwan dataset. Additionally, it achieved an average mAP@50 of 87%, mAP@50-95 of 48%, precision of 83.9%, recall of 70.3%, and F1-score of 76.4% on the TLDDCV dataset. These results demonstrate the generalizability and effectiveness of the proposed system in real-world agricultural situations.
Reproduction assets foundThe paper uses two public Roboflow tomato leaf disease image datasets and states its source code is publicly available on Zenodo, all with explicit availability statements and URLs.Dataset · publicThe first dataset is the Taiwan dataset, which can be found at the following link: (https://universe.roboflow.com/bryan-b56jm/tomato-leaf-disease-ssoha).Open asset ↗tomato-leaf-disease-ssohalines:317-328Dataset · publicThe second dataset is the TLDDCV dataset, which can be found at the following link: (https://universe.roboflow.com/sylhet-agricultural-university/tomato-leaf-diseases-detect)Open asset ↗tomato-leaf-diseases-detectlines:317-328Code · publicThe source code of the proposed framework, including the implementation of the proposed methodology and experimental setup, is publicly available in the Zenodo repository: https://doi.org/10.5281/zenodo.20777853 .Open asset ↗Zenodo · 10.5281/zenodo.20777853lines:317-328Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published8 Aug 2026International Journal of Engineering Trends and TechnologyCited by 0 · OpenAlex ↗
Brassica juncea (Mustard) is one of the most economic seed vegetable crops of the world, playing a major role in the production of world edible oil and the agricultural economy. The third-largest producer is India, which had an area under cultivation of about 8.6 million hectares of Mustard in 2021 22, and annual revenue of over USD 5 billion in 2021 22. Nevertheless, the presence of diseases like Alternaria Leaf Spot, White Rust, Powdery Mildew, and Septoria Leaf Spot threatens yield and quality by up to an estimated 2070% loss every year, based on the severity of the disease, and thus economic losses are estimated at over USD 1.5 billion per year in India alone. Traditional diagnostic systems are based on a manual examination of an agronomist trained to look at the sample and make a judgment, which is time-consuming, subjective, and subject to human error. Current deep learning methods of automated disease detection, promising as they are, are prone to inaccuracies on complex disease patterns, poor uncertainty estimation that is essential in real-world implementation, and poor generalizability to different field conditions. In response to these drawbacks, Swin-BNN-RF, a hybrid framework that combines Swin Transformer as a hierarchical attention-based feature extractor, Bayesian Neural Network (BNN) with symmetrized posterior as a probabilistic learner, and a Random Forest (RF) as an ensemble classifier, is proposed in this study. The Swin Transformer also harnesses local and global spatial biases with its shifted window self-attention network, and it is able to extract better features on leaf images of high-resolution. The uncertainty estimates of the BNN component are trusted, and unambiguous predictions are highlighted to get the opinion of the human expert. Random Forest classifier uses the bagging and boosting ensemble methods to improve stability and the robustness of the classification. A large dataset was experimented with; it consisted of more than 10,000 samples per category of disease in four diseases. In the case of binary classification, the proposed model was 98.32% accurate, 98.52% precise, 98.70% recall, and 98.36% F1. On multi-class classification, it obtained 97.50, 97.82, 98.51, and 97.46 accuracy, precision, recall, and F1 score, respectively, which showed consistent performance in comparison with state-of-the-art models such as EfficientNet, MobileNet, and Residual Networks. The contribution of each component is verified by the Ablation studies and statistical analysis of significance (p
Accurate detection of small-target diseases on apple leaves is of great importance for optimizing orchard management and facilitating precision agriculture. Small-target disease detection remains challenging due to insufficient features and complex backgrounds. This work proposes an effective detector for apple leaf small-target diseases called ALSDet. The global context module is integrated into Stage2 to Stage4 of the ResNet-50 backbone, yielding a refinement of the feature-extraction architecture. In the bottleneck blocks of the Stage2 and Stage3, dilated convolution is used in place of the normal 3×3 convolution to enlarge the receptive field for small targets and strengthen feature extraction. During the model training phase, a multi-scale training strategy combined with the online hard example mining method is adopted to focus on learning hard samples and enhance adaptability for various scale targets. According to the experimental results, ALSDet obtains a mean average precision (mAP) of 65.6% and an average recall (AR) of 71.2% on the dataset. The proposed model achieves the highest levels in both mAP and AR when compared to the popular object detection models, such as Cascade R-CNN, Faster R-CNN, GFL, Grid R-CNN, Libra R-CNN, FCOS, VFNet, RetinaNet, SSD, YOLOv7, and YOLOv8. For small-target diseases like rust and frog eye leaf spot, the average precision surpasses 87% with an intersection over union (IoU) threshold of 0.5. These results confirm that ALSDet achieves stable performance against existing methods, demonstrating its potential as a practical tool for intelligent orchard disease management.
Identifying plant health conditions is an emerging precision-agriculture and food-security challenge, intensified by deploying deep-learning models on memory- and power-constrained edge devices. We present a TinyMLOps pipeline spanning model design, optimization, quantization, and deployment across diverse edge devices, evaluated under controlled laboratory conditions. Using a dataset derived from the PlantVillage benchmark, 39 fine-grained classes are aggregated into three superclasses: healthy leaf, unhealthy leaf, and no leaves. The resulting system therefore performs plant health-status classification and background filtering rather than diagnosing specific diseases. We train a MobileNet-based convolutional neural network jointly optimized for classification accuracy and computational efficiency, adopting state-of-the-art hyperparameter optimization (HPO) tools. Five models are selected, four from the Pareto Front and one as the biggest evaluated model during HPO, converted to LiteRT and ONNX, and evaluated at float32 and post-training int8 precision on a Raspberry Pi Zero 2 W and an STM32H743ZI microcontroller. At float32, LiteRT is 1.87–2.65× faster than ONNX Runtime on the Raspberry Pi across all five models. Relative to their float32 LiteRT counterparts, the int8 LiteRT models are 2.83–3.67× smaller on disk and 21.3–31.2% faster on the same board, at a cost in F1-score of between 0.0010 and 0.0068. On the microcontroller, only the two smallest models deploy at both precisions; for these, the fully quantized int8 variants are 4.3× faster and 3.85× smaller in MCU flash footprint than the float32 counterparts. The mid-range model fits the 2 MB flash and 1 MB RAM budget only when quantized, while the two largest models exceed it in every configuration tested.
Reproduction assets foundThe paper's plant-phenotyping measurements are based on a derived PlantVillage dataset (39 classes aggregated into three superclasses) that the authors explicitly state is openly available in their own GitHub repository, also catalogued in the AgrifoodTEF Data Space. No author analysis code or trained model checkpointsDataset · publicThe data used in this study are derived from the openly available GitHub repository available at https://github.com/FBK-OpenIoT/PlantVillage-AugNoLeaves , accessed on 1 July 2026.Open asset ↗FBK-OpenIoT/PlantVillage-AugNoLeaveslines:1355-1403Dataset · publicPlantVillage-AugNoLeaves—AgrifoodTEF Data Space Catalogue. 2025. [(accessed on 1 July 2026)]. Available online: https://dataspace.agrifoodtef.eu/asset/did:op:3091fdcf83a05784416e585e4e45ea24afaf8c925a891921a3949acd98426f65Open asset ↗did:op:3091fdcf83a05784416e585e4e45ea24afaf8c925a891921a3949acd98426f65lines:1424-1474Plant phenotyping relevance matchOpenAlex · checked 14 Sept 2026
Monitoring continuous agricultural canopies is fundamentally limited by the geometric constraints and computational bottlenecks of traditional 3D reconstruction. This study presents a 3D volumetric phenotyping pipeline integrating semantic mask generation with 3D Gaussian Splatting (3DGS) to quantify greenhouse cucumber canopy architecture. To drive component-specific optimization, we evaluated custom-trained convolutional networks (YOLO11) against a zero-shot foundation model (SAM3), determining that SAM3 provided the necessary boundary precision for accurate spatial isolation. The optimized 3DGS model outperformed implicit NeRF baselines, preserving fine-scale morphological details at real-time rendering speeds ( ≈ 48 FPS). To enable actionable measurement, a uniform voxelization protocol was applied to the point cloud, successfully neutralizing algorithmic densification bias. This technical framework yielded highly accurate physical geometry, achieving a Root Mean Square Error (RMSE) of ≤ 0.59 cm against in situ leaf measurements. Transitioning to agronomic interpretation, the pipeline was deployed to quantify complex canopy architecture. It mathematically mapped structural congestion zones and provided a temporal validation of a standard pruning intervention, explicitly capturing the geometric increase in lower-canopy porosity and the upward translation of biomass. This framework provides a robust, scale-accurate tool for monitoring plant architecture and guiding dynamic canopy management.
Early and accurate plant disease detection is essential for reducing crop losses and supporting sustainable agricultural management. Although deep learning-based approaches have achieved strong performance in plant disease analysis, many existing models require substantial computational resources, which limits their use in resource-constrained agricultural environments. This study proposes ShuffleNetV2-hSimKD, a lightweight integration framework for plant disease detection. It adopts ShuffleNetV2 as the backbone and incorporates the parameter-free SimAM attention mechanism to enhance disease-related feature representation without introducing additional learnable parameters. In addition, the standard ReLU activation function is replaced with h-swish to improve nonlinear feature extraction and preserve informative feature responses. A hybrid knowledge distillation strategy is further employed to transfer both output-level and feature-level knowledge from a high-capacity teacher model to the lightweight student network during training. Unlike previous studies that apply these techniques in isolation, ShuffleNetV2-hSimKD synergistically integrates parameter-free SimAM, h-swish optimization, and hybrid KD to overcome the representation limitations of lightweight backbones in subtle disease symptom detection. The proposed framework was evaluated on a balanced subset of the PlantVillage dataset, in which leaf images were categorized as healthy or diseased. ShuffleNetV2-hSimKD achieved an accuracy of 90.41% with only 1.4M parameters and 151M FLOPs. Compared with representative lightweight Convolutional Neural Networks (CNNs), the proposed model achieved improved accuracy and recall while maintaining low computational complexity. These results demonstrate that ShuffleNetV2-hSimKD provides an effective balance between detection performance and computational efficiency, highlighting its potential as a lightweight candidate for plant disease detection in resource-constrained agricultural scenarios.
Introduction The precise detection of microscopic tea leaf diseases is a prerequisite for sustainable precision agriculture. While recent deep learning advancements often favor architectural complexity, this "complexity bias" frequently introduces computational redundancy-a "complexity tax"-that destabilizes gradient flow and fails to resolve critical resolution bottlenecks for micro-lesion identification. Methods We propose Opti-YOLOv11n, a minimalist optimization paradigm prioritizing physical input fidelity and stabilized gradient dynamics. Utilizing a dataset of six pathological categories, our framework employs high-resolution scaling (832×832) combined with a momentum-based SGD optimizer and a cosine annealing schedule to reconstruct essential spatial textures. Results Opti-YOLOv11n achieved a peak Precision of 98.87% and a Recall of 95.97%, while reducing the parameter count to 2.35 M-a 9.2% decrease relative to the baseline-and maintaining a real-time inference speed of 104.7 FPS on edge-simulated hardware. Discussion Statistical verification via 5-fold cross-validation confirms superior generalization stability. These results substantiate that strategic structural pruning and physical input scaling provide a more robust technical benchmark for autonomous plant protection than the adoption of excessive architectural depth.
Leaf chlorophyll content (LCC) is a key indicator for assessing the photosynthetic capacity and nutritional status of winter wheat. Among traditional LCC estimation methods, empirical models lack a physical basis and have poor generalisability, while physical models are widely applicable but suffer from ill-posed inversion problems. Hybrid inversion methods, which integrate radiation transfer models such as PROSAIL with machine learning, offer both the interpretability of physical models and the efficiency of machine learning; however, they are still affected by the domain shift between simulated and measured data, which limits their generalisation performance. Transfer Component Analysis (TCA), a domain adaption method, can effectively alleviate this problem. In this study, hyperspectral and LCC data were collected in the field, and simulated data were generated using the PROSAIL model; a sensitivity analysis was conducted to identify LCC-sensitive bands. A genetic algorithm was applied to the measured data for band selection and, together with the results of the sensitivity analysis, yielded an optimal set of 30 characteristic bands for subsequent modelling. Three datasets were constructed: measured data only, a direct mixture of measured and simulated data, and a TCA-fused mixture of measured and simulated data. Four models-gradient boosting regression (GBR), random forest (RF), support vector regression (SVR) and deep neural network (DNN)-were developed for each dataset. The results show that: (1) the LCC-sensitive bands are concentrated in the 450-660 nm and 680-720 nm ranges; (2) the model built on the TCA-fused data (R² = 0.722, RMSE = 6.792) outperformed those built on the measured-only data (R² = 0.682, RMSE = 7.259) and the directly mixed data (R² = 0.616, RMSE = 7.976); (3) for the TCA-fused data, the four models differed considerably in accuracy, with SVR performing best (R² = 0.723, RMSE = 5.363), followed by RF (R² = 0.630, RMSE = 6.201) and GBR (R² = 0.575, RMSE = 6.650), whereas the DNN performed worst (R² = 0.388, RMSE = 7.947), probably owing to the limited sample size.
Jiří Chuchlík · Jan Stejskal · Jaroslav Čepl · Jiří Korecký · Salvador A. Gezan · Yousry A. El‐Kassaby
Field / plotRGB / grayscaleLeafSegmentationGrowth / development / phenology
Bud flushing timing is a crucial attribute for forest geneticists and physiologists studying adaptive variation in tree species. In clonal populations such as Norway spruce seed orchards, accurate assessment of bud flushing is essential, as this attribute strongly influences survival and fitness. However, traditional bud-flushing assessment relies on visual scoring scales that are inherently subjective and susceptible to observer bias, leading to inconsistencies when multiple technicians collect data. To address these limitations, we introduce a novel quantitative phenotype derived from segmented RGB images for quantifying bud flushing that objectively measures the ratio of light-green current-year needles to the total visible needle area of each grafted ramet. By extracting and summarizing digital image data for each ramet, this approach provides a standardized and reproducible assessment of bud flushing phenology. We evaluated clonally replicated ramets in two sites across two seasons and compared this image-derived flushing ratio with conventional visual scoring and color indices. Using mixed models, we quantified genetic control and cross-orchard stability of clonal performance. The image-derived flushing ratio tracked visual-score progression but showed higher broad-sense heritability (0.54–0.70) than visual scoring (0.45–0.65) and canopy color indices (0.20–0.51). Although site means differed strongly, clonal ranking remained stable between orchards for the image-derived flushing ratio and for visual scoring, whereas color indices showed lower stability and greater uncertainty. Overall, these results support the image-derived flushing ratio as a structure-based phenotype that improves the precision and cross-site transferability of genetic evaluation of bud flushing in Norway spruce seed orchards.
Precise and reliable diagnosis of leaf diseases in tomato is essential for enhancing crop cultivation and minimizing agricultural losses. While deep learning models have performed well on benchmark datasets, the majority of present techniques rely on flat multi-class classification, which predicts all disease categories simultaneously. Such formulations promotes inter-class confusion, particularly when biologically different diseases with similar visual symptoms are learned within a same model. To overcome this constraint, we propose a biologically structured hierarchical deep learning framework in this study. Instead of directly classifying 10 disease classes, the proposed method first classifies leaf images into meaningful biological groups such as bacterial, fungal, pest-associated and healthy using a vision transformer (ViT) model. Then, specialized convolutional neural network (CNN) experts perform fine-grained classification within each category. The proposed hierarchical model shows an overall accuracy of 97.8%, when validated on PlantVillage tomato dataset. A flat ViT model trained with class-weighted loss obtained 96.3% accuracy, whereas a flat CNN model reached 99.3% under clean conditions but decreased sharply to 41% under Gaussian perturbation ( σ = 0.05). On the other hand, the hierarchical model performed steadily under noise with 97.4% accuracy at the same perturbation level. These results indicate that adding biological structure to model design reduces confusion, helps prevent imbalance effects and increases robustness, providing a more trustworthy and interpretable solution for real-world agricultural disease diagnosis.
Reproduction assets foundThe paper's phenotyping/classification measurements are based on the public PlantVillage tomato leaf image dataset, which the authors explicitly state was analyzed and provide a public Kaggle URL. No author analysis code, trained models, or paper-specific supplementary assets are described with availability language.Dataset · publicIsabel Luna-Maldonado , Autonomous University of Nuevo León, Mexico
Reviewed by: Noredine Hajraoui , Moulay Ismail University, Morocco
Tri Handhika , Universitas Gunadarma Pusat Studi Komputasi Matematika, Indonesia
Data availability statement
Publicly available datasets were analyzed in this study. This data can be found here: https://www.kaggle.com/datasets/charuchaudhry/plantvillage-tomato-leaf-dataset .
Author contributions
HG: Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. SR: Supervision, Validation, Writing – review & editing. BL: Supervision, Validation, Writing – review & editing.
Conflict of interest
The author(s) declared thaOpen asset ↗Kaggle · charuchaudhry/plantvillage-tomato-leaf-datasetlines:588-616Code / dataset availability confirmedEurope PMC · checked 5 Sept 2026
TomatoLiDAR / point cloudLeafMorphology / geometry measurementSegmentationLeaf traits
Leaf parameters are crucial indicators reflecting the growing status of plants. Monitoring and analysis of leaf parameters significantly contributes to the improvement of crop yield and food quality. This study focused on three tomato plant varieties commonly grown in the Netherlands and proposed a fully automatic pipeline for leaf phenotyping. Three-dimensional (3D) point clouds of target plants were acquired with a specially designed imaging unit naming Maxi-Marvin. A semantic segmentation of plant organs was performed with PointNet++ model. To mitigate point cloud resolution decrement, the down-sampling operation in the baseline model was replaced with a distributed segmentation strategy. Leaf instances were further identified with Density-Based Spatial Clustering of Applications with Noise (DBSCAN), followed by a morphological phenotypic trait quantification based on 3D geometrical analysis. Target phenotypic traits including leaf length, leaf width, and leaf area. The evaluation results indicated that the distributed segmentation strategy achieved the best F 1 scores of 0.98 with block size set to 30,000. The Mean Average Errors (MAE) of leaf length, leaf width, and leaf area estimation were 2.09 cm, 1.78 cm and 8.98 cm 2 respectively. The estimation accuracies for leaf length, leaf width, and leaf area were 91.98%, 92.66%, and 89.67%, respectively.
Reproduction assets foundThe paper's tomato point cloud dataset (with semantic and leaf instance annotations used for the phenotyping pipeline) is publicly available on Kaggle via a footnote. NPEC website is a facility page, and Open3D is a generic library, so neither qualifies.Dataset · public2. ^ The dataset used in this study is available at: https://www.kaggle.com/datasets/xinbolai/vtc-tomatoOpen asset ↗Kaggle · xinbolai/vtc-tomatolines:545-624Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published4 Aug 2026International Journal of Innovative Science and Research TechnologyCited by 0 · OpenAlex ↗
Accurate, field-deployable diagnostic tools are needed to close the diagnostic gap that limits fungicide targeting among smallholder tomato farmers in Northern Nigeria. This study developed and validated a lightweight convolutional neural network (CNN) for automated diagnosis of five major tomato fungal diseases plus three additional common conditions, trained on field-collected leaf images from Kano and Kaduna States. A MobileNetV2 architecture pre-trained on ImageNet was fine-tuned via transfer learning on more than 10,000 images across ten disease and health classes, using farm-level dataset splitting to prevent data leakage and five-fold cross-validation for model selection.
Abstract Background Leaves maintain hydraulic homeostasis during photosynthesis through the coordinated action of stomata, which regulate gas exchange and transpiration, and veins, which supply water to the leaf lamina. While functional links between stomatal and vascular traits are known in dicots, their potential genetic coordination in C4 crops remains poorly understood. We investigated the genetic architecture of these traits in maize using a Multi-parent Advanced Generation Inter-Cross (MAGIC) population and a low-cost, high-throughput phenotyping platform integrating leaf clearing, digital microscopy, artificial intelligence, and image analysis Results We phenotyped 285 recombinant inbred lines and the MAGIC founder lines, generating 8,072 images from 2,026 leaf samples taken from seedlings grown in controlled conditions. A YOLOv8-based model automatically detected stomata, while a custom and efficient image-processing pipeline quantified vein traits and stomatal spatial distribution patterns along cell bundles. This enabled simultaneous characterization of stomatal density, size, and distribution together with vein density, thickness, and bundle-associated spatial patterning. Substantial phenotypic variation was observed among genotypes, with strong correlations between abaxial and adaxial traits but no significant correlations between stomatal and vein traits. QTL mapping identified 37 genomic regions associated with stomatal and vein traits, including loci containing known developmental regulators such as stomatal density and distribution1 and stomagen1 , as well as novel loci controlling stomatal spatial patterns, divergence between leaf surfaces and veins traits. Conclusions These results support independent genetic control of stomata and veins and decoupled contribution to water-use efficiency, providing a novel genetic framework to independently optimize leaf hydraulic capacity and gas exchange in target environments.
Purpose Conventional plant disease detection is time-consuming and prone to human error. The purpose of this study is to propose an edge artificial intelligence (AI)-based deep learning framework for plant disease detection under real-field conditions. The model integrates convolutional neural networks (CNNs) with a Sliding Window Mean Absolute Deviation (SWMAD) preprocessing technique to address illumination variability and complex background conditions. Design/methodology/approach A dual-layer CNN model is developed for plant disease classification using HSV segmentation, flood-fill segmentation and SWMAD preprocessing to handle real-field variations. The model is trained on PlantVillage and real farm images and deployed via TensorFlow Lite for real-time offline detection. TensorFlow Lite is used to enable efficient on-device inference for deployment in resource-constrained environments. Findings Benchmark experiments conducted on the PlantVillage dataset achieved a classification accuracy of 99.91% under controlled conditions. On the hybrid dataset comprising PlantVillage and real-field images, the optimized CNN framework achieved a validation accuracy of 95.01% following extensive evaluation of optimizers, layer architectures and worker configurations. The integration of the proposed SWMAD preprocessing technique into the finalized architecture further improved the validation accuracy to 97.38%, demonstrating enhanced robustness and classification performance under practical agricultural conditions in the final deployed model. Originality/value The originality of this study lies in several novel contributions. First, we introduce an SWMAD-based preprocessing technique, which enhances local statistical variations in leaf images by capturing pixel-level deviations from neighborhood intensity means. Unlike, conventional preprocessing methods, SWMAD is specifically designed to handle real-field challenges such as illumination variation, noise and complex backgrounds. The improvement in validation accuracy demonstrates the effectiveness of the proposed approach in capturing more discriminative features compared to existing techniques, thereby improving overall model robustness and reliability in practical agricultural environments.
This paper explores how deep learning methods can be used to monitor the health of crops and identify diseases, particularly for the apple crop. As the need for food security and sustainable farming methods increases, there is a strong demand for early detection of crop diseases. We have used a Convolutional Neural Network (CNN), based on the model of VGG16 architecture, since the model is known to be effective in image classification. The dataset contained 7771 training images for to enhance machines deep learning regarding plant diseases. Following that, validation image collection of 1747 images divided into four health conditions of the apple crops. In addition, the model was evaluated using 196 new images as a final test. To enhance the model capacity for recognizing diseases of real leaves rather than just memorize exact training pictures, data augmentation was used with ImageDataGenerator of TensorFlow. This means the training images were zoomed, rotated, and shifted to enable the machine detects more variations. Ten epochs of training were performed to measure the model accuracy. The results indicated that the model obtained significant improvement in training and validation accuracy from 56.43% to 78.12% and 92.94 to 96.93%, respectively. Most impressively, the final test dataset, which contained completely new images, scored an accuracy rate of 98%. The results indicate that in the architecture field the application of deep learning methodologies is effective, suggesting that automated detection of diseases by using sophisticated image analysis manages crop diseases identification efficiently. Combination of these methods successfully creates avenues for novel research to built real-time systems of crop disease monitoring, which help farmers increase their productions and farm their lands sustainably.
The simultaneous diagnosis of diseases and evaluation of age quality grades in tea leaves are critical for precision agriculture and the economic valuation of tea products. Although deep learning has shown promise in agricultural vision tasks, current multi-task models often suffer from performance degradation due to feature conflicts: tea leaf disease recognition relies heavily on macro-structural lesions, whereas tea leaf-age quality grading depends on micro-textural features such as trichome density and color uniformity. To address this discrepancy, we propose a novel dual branch fusion network. Our architecture fundamentally decouples the feature extraction process by utilizing a dual branch mechanism. The first branch employs global average pooling to capture first-order spatial statistics; it can retain the global structural layout necessary for macro-lesion detection. The second branch introduces a dimensionality-reduced self-bilinear pooling module to compute second-order covariance matrices; it can effectively capture the fine-grained textural patterns essential for micro-grade classification. These decoupled features are subsequently fused and optimized through a weighted multi-task loss function. Experimental results on a comprehensive tea leaf dataset demonstrate that the proposed dual fusion framework significantly outperforms baseline models. The proposed network can rescue the disease classification accuracy drop observed in standard bilinear models while maintaining exceptional grading performance. Furthermore, the proposed network maintains a compact parameter footprint and low computational complexity. This balance renders it suitable for deployment on agricultural Internet of Things edge devices where inference speed is critical.
Reproduction assets foundThe paper's Data Availability Statement publicly releases the two tea leaf image datasets used for its phenotyping tasks (disease recognition and leaf-age quality grading) via Mendeley Data. The authors' analysis code and trained models are only promised 'upon acceptance' with no public URL, so they do not qualify.Dataset · publicThe tea leaf disease recognition dataset analyzed in this study is available from https://data.mendeley.com/datasets/744vznw5k2/3 (accessed on 11 February 2026)Open asset ↗744vznw5k2/3lines:514-565Dataset · publicthe tea leaf grading dataset is available from https://data.mendeley.com/datasets/7t964jmmy3/1 (accessed on 11 February 2026)Open asset ↗7t964jmmy3/1lines:514-565Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Published4 Aug 2026Journal of Agriculture and Ecology Research InternationalCited by 0 · OpenAlex ↗
Crop stress develops through a sequence that begins with molecular and biophysical perturbation, progresses through physiological dysfunction, and only later becomes visually apparent. Precision agriculture therefore requires sensors that can shorten the interval between stress onset and actionable diagnosis while preserving spatial context. This critical narrative review examines the complementary roles of nanosensors, plant-wearable and implantable electronics, proximal sensing, unmanned aerial vehicles, satellite remote sensing, and geographic information systems in early crop-stress detection. Literature published from 2000 to 1 June 2026 was selected through live searches of accessible scholarly indexes, DOI registries, publisher records, institutional repositories, and citation networks, with foundational studies retained where necessary. The evidence shows that nano-enabled interfaces can measure early biochemical, ionic, volatile, electrical, and microclimatic signals at high temporal resolution, whereas geospatial technologies reveal the distribution, persistence, and management relevance of stress across canopies and fields. Optical nanotube sensors, surface-enhanced Raman probes, electrochemical microneedles, ion-selective wearables, and flexible leaf sensors have demonstrated biologically meaningful signals before visible symptoms in controlled or pilot field settings. Yet most remain constrained by sparse sampling, crop-specific calibration, bio-interface effects, power and communication burdens, uncertain durability, and limited agronomic validation. Geospatial methods are operationally more mature, particularly thermal and multispectral imaging for water stress and hyperspectral imaging for disease and nutrient-related changes, but they often infer stress through non-specific proxies that are confounded by canopy structure, atmosphere, soil background, phenology, and co-occurring stresses. The strongest future architecture is therefore not a contest between nanoscale and landscape-scale sensing. It is a multiscale system in which physiologically specific plant sensors anchor and interpret spatial imagery, while remote sensing directs where high-specificity measurements and interventions are most valuable. Progress depends on prospective field trials, reference measurements, uncertainty-aware data fusion, interoperability, lifecycle safety assessment, and decision thresholds linked to economic and agronomic outcomes.
Reliable traits are needed for identification of tea ( Camellia sinensis ) cultivars, yet the stability of leaf morphology and color across leaf positions remains unclear. This study evaluated inter-cultivar variation and positional stability in leaf morphological, RGB color, and SPAD traits in six predominant cultivars. One-year-old shoots were sampled in a completely randomized design, and five fully expanded leaves below the apical bud were analyzed. SPAD values were measured with a chlorophyll meter, and scanned images were used to extract contour and RGB traits. Data were analyzed using ANOVA, correlation analysis, PCA, and discriminant analysis. Leaf morphology differed among cultivars and leaf positions, with significant cultivar-by-position interactions; however, the width-to-length ratio differed among cultivars but remained stable across positions in these cultivars. SPAD values increased with leaf position and were strongly associated with RGB components, being negatively correlated with R and G and positively correlated with B. Morphological traits explained 52.988% of total variance in PCA and yielded 64.6% overall classification accuracy, with LaoHan showing the highest accuracy (83.3%). Misclassification was concentrated among genetically similar cultivars. These findings suggest that stable leaf shape proportions and SPAD-RGB relationships provide useful descriptors, whereas genetic relatedness limits morphology-based cultivar identification under the present conditions.
Reproduction assets found保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。Dataset · publicThe following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/biology15151283/s1 , Table S1: Original data of leaf morphological traits, RGB values, and SPAD values from six tea cultivars in this study.Open asset ↗lines:368-409Plant phenotyping relevance matchEurope PMC · bioRxiv · checked 15 Sept 2026
1 Summary Embolism, the formation of air bubbles in the plant water transport system, is a mechanistic driver of plant death. The Optical Vulnerability Technique (OVT) is an imaging method for non-invasive quantification of embolism (including P50, a common metric for drought vulnerability), which can also provide detailed spatial and temporal information. Its major cost lies in the post-processing of thousands of images. Here we designed, tested, trained, and make publicly available a neural network model to automate post-processing of OVT images. Using a dataset of 65 leaves from Senecio pterophorous , we compared our model predictions to results obtained via traditional post-processing by an expert. Our model resolved P50 to within 0.027 MPa of the expert-processed data with training taking 30 minutes to 2.5 hours and model-runtime in the order of seconds to minutes, demonstrating its promise for increasing the efficiency and throughput of P50 calculation. The model’s performance in replicating the pixels that constitute embolism events was lower (mean event-frame IoU of 0.38). We invite the community to utilise our model but emphasise that it does not replace the expert-processing pipeline and that care must be taken when considering applying this and similar approaches to OVT data.
Background Accurate and efficient detection of citrus leaf diseases is important for orchard monitoring, early intervention, and intelligent disease management. However, practical application in natural orchard environments remains challenging because of complex backgrounds, large variation in symptom scale, strong interclass similarity, and limited edge computing resources. Methods In this study, a lightweight object detection model, SKM-YOLOv11, was developed based on YOLOv11n for citrus leaf disease recognition in natural scenes. A self-built dataset was constructed from public images and field images, containing 6414 images across six classes. StarNet-S050 was introduced as the backbone, C3k2-Star was designed to enhance feature fusion across scales, and a lightweight shared detection head, MNS-Head, was constructed to reduce prediction redundancy. Results Compared with YOLOv11n, SKM-YOLOv11 increased Precision, Recall, and mAP @0.5 by 2.92, 1.47, and 0.95 percentage points, respectively. Meanwhile, FLOPs, parameter count, and model size were reduced by 31.7%, 32.7%, and 34.55%, respectively. Edge deployment on Jetson Orin NX Super achieved an inference speed of 72.45 frames per second. Conclusion The proposed model has strong potential for real-time citrus disease screening on resource-constrained devices and provides a feasible solution for edge-based intelligent disease management in natural orchards.
Soybean diseases caused by fungal, bacterial, and viral pathogens represent a major constraint to global agricultural productivity. Although molecular phylogenetic analyses have advanced the understanding of pathogen evolution, the extent to which disease phenotypes reflect evolutionary relationships remains poorly understood. In this study, we developed an integrative framework combining deep learning-based phenotypic analysis with phylogenetic inference to investigate the relationship between soybean disease symptoms and pathogen evolution. An EfficientNet-B0 convolutional neural network (CNN) was trained to classify 10 soybean disease classes comprising 703 leaf images and achieved a mean cross-validation accuracy of 98.72 ± 1.17%, a weighted F1-score of 98.74 ± 1.16%, and a macro F1-score of 98.47 ± 1.74%. Evaluation on a held-out test set generated through image-level partitioning yielded an accuracy of 96.19%, a weighted F1-score of 96.28%, and a macro F1-score of 95.86%. Latent feature embeddings revealed a structured phenotypic space with clear separation among most disease classes and enabled quantitative analyses of phenotypic similarity. To provide biological context, taxonomy-derived distance matrices and sequence-based phylogenetic analyses of the fungal subset using 28S rRNA sequences were compared with CNN-derived phenotypic representations. A Mantel test identified a moderate and statistically significant association between phenotypic and phylogenetic distances (Spearman r = 0.3393, p = 0.0050), indicating that pathogen evolutionary history contributes to disease phenotype while explaining only part of the observed phenotypic variation. Overall, the results demonstrate that deep learning effectively captures biologically meaningful phenotypic information while highlighting that disease symptoms arise from the combined influence of pathogen evolution, host responses, and environmental conditions. This study provides an integrative framework for combining image-based phenotyping with phylogenetic analysis to support biologically informed interpretation of plant disease phenotypes.
Abstract Leaf color is an important trait affecting vegetable quality, yield, and market value. However, traditional methods for leaf color assessment are often subjective or destructive, which limits accurate and high-throughput phenotyping. In this study, an unmanned aerial vehicle (UAV)–based multispectral imaging platform was used to collect phenotypic data from 214 Chinese cabbage inbred lines at the rosette stage. A multispectral UNet model was applied to segment individual plants, and a membership function was used to quantify leaf color on a continuous scale. Based on these high-throughput phenotypic data, a genome-wide association study was used to identify two candidate genes, BrEMB976 and BrGSH2, on chromosome A06. Subsequent virus-induced gene silencing analysis showed that silencing these genes altered leaf color. In addition, a deep learning-based genomic selection model, BrDeepGS, was developed for leaf color prediction, which achieved a Pearson correlation coefficient of 0.853. These results demonstrate the potential of integrating UAV-based high-throughput phenotyping, candidate gene analysis, and genomic prediction for leaf color evaluation and selection in Chinese cabbage breeding.
Annotation scarcity, poor model generalization and lagged data processing remain key bottlenecks hindering the practical deployment of phenotyping robots. To address these issues, we developed a novel phenotyping robot capable of online 3D reconstruction and zero-shot segmentation directly on the edge. Diverging from conventional semantic SLAM, our core contribution is RT-ZSDR, a framework featuring two key methodological novelties. First, we introduce the ForeCut pipeline for target extraction, which innovatively fuses DINO features with 3D geometric spatial information, leveraging multi-view semantic-spatial consistency to achieve annotation-free, zero-shot dense segmentation and reconstruction. Second, we designed a hardware-coupled loop closure strategy utilizing the robotic arm's kinematic feedback as prior constraints to significantly improve loop closure recall. Supported by edge computing Jetson Orin NX, the tracking and segmentation process takes approximately 0.24 s per frame after an initialization period of 1.82 s. RT-ZSDR's phenotypic measurements demonstrated strong correlations with reference baseline in both laboratory settings (n=90, PlantEye measurements as reference baseline; R 2 =0.990, 0.939, 0.725, and 0.861 for plant height, projected leaf area, surface area, and volume) and practical greenhouse environments (n=48, manual measurements as reference baseline; R 2 =0.965, 0.862 for plant height and stem diameter). Additionally, evaluated against COLMAP benchmarks (n=24), the system achieved a mean 3D reconstruction F1-score of 0.816.
As a critical structural component that connects almost all other types of plant organs, the stem system not only supports the weight of the total plant, but also serves as a vital channel for nutriment transportation. Accurate phenotypic measurement of stem instances is of practical significance for assessing crop growth dynamics and predicting yield. To address current 3D phenotyping challenges of crops such as the difficulty in separating stem segments from the stem system and the low accuracy in stem length measurement, we propose a Hierarchical-annotation-guided Phenotypic Framework (HagPF) for Stem Instance Segmentation and Length Measurement in plant point clouds. Specifically, a hierarchical leaf-stem organ instance annotation strategy is devised to effectively train a PSegNet network for leaf and stem instance segmentation. The segmentation is then followed by a shape-adaptive measurement algorithm to automatically measure the length of stem segments that are morphologically diverse in space. On a 3D dataset comprising four crop species, the proposed framework achieved an Intersection over Union (IoU) of 95.45% for organ semantic segmentation and a Mean Weighted Coverage (mWCov) of 87.87% for instance segmentation (both stem and leaf). Regarding to the stem length measurement, the method obtained an average Root Mean Square Error (RMSE) of 1.044 cm and a relative error of 11.907%, outperforming 7 mainstream methods. The relevant dataset and source code can be found at: https://github.com/Jinx00/stem-length-measurement .
Mallavarapu A, Merriam I, Riemann M, Hu EY, Cai L, Helliker B, Turner KT, Kagan CR.
LeafStress / disease detectionWater status / transpiration
We report the design and assembly of geometrically and compositionally complex dielectric metasurfaces, termed nanocrystal-adaptive polymer composite optical metasurfaces (NCOMs), architected from nanocrystal (NC) building blocks and stimuli-responsive polymers to enable high figure-of-merit optical leaf sensors. These NCOM sensors are fabricated at scale by direct nanoimprint lithography into UV-curable TiO2 NC inks to structure optical metasurfaces in the geometry of surface relief gratings atop waveguides that support high reflectivity, narrowband guided-mode resonances. A room-temperature ligand exchange process creates a nanoporous TiO2 scaffold that is subsequently infiltrated with a stimuli-responsive polymer while retaining the metasurface geometry and optical quality. As a proof of concept, we demonstrate optical humidity sensors by incorporating a moisture-responsive biopolymer chitosan into the nanoporous TiO2 metasurfaces, resulting in a 430% improvement in sensitivity compared to conventional refractive-index sensors where the polymer is adjacent to the metasurface. The NCOM geometry is engineered to position its resonance at wavelengths above 750 nm, enabling detection with standard VIS-NIR Si photodetectors while maintaining broadband transparency across the photosynthetically active 400-700 nm wavelength range for unobstructive monitoring of crop health. NCOM sensors are mounted face-down on leaves, and the highly reflective resonances are readily distinguished from the leaf background and respond dynamically to leaf water stress, enabling passive, battery-free monitoring of leaf surface humidity.
Advancements in stay-green phenotyping are increasingly utilizing hyperspectral sensing technology to assess crop response under extreme environmental conditions. Yet, the effectiveness of different spectral features in explaining stay green remains to be fully elucidated. This includes identifying which bands and spectral indices are more effective in capturing the genotypic differences in stay-green traits. The main objective of this study was to evaluate hyperspectral leaf reflectance as a means to estimate stay-green visual scores (SGVS) as an indicator of drought tolerance and to further understand whether chlorophyll absorption-band spectral indices can differentiate SGVS classifications during post-flowering stages of maize. The experiment was conducted over two growing seasons in Germany, comprising 18 maize genotypes under two contrasting water availability conditions. We measured leaf hyperspectral reflectance using a spectroradiometer in the second, fourth, and sixth week after flowering, along with stay-green traits measurements. We employed raw spectral reflectance, hyperspectral vegetation indices (VIs) in combination with random forest (RF) and ANN models to predict SGVS. Results showed that drought stress significantly affected stay-green-related traits and led to a 43.5% decrease in grain yield in the inbred lines. The grain dry yield (GDY) was positively correlated with stay-green visual scores (SGVS), with higher SGVS associated with higher GDY. Stay-green traits were correlated with various VIs, with the best correlation observed for the Chl_NDI (r = 0.91). Stay-green groups were successfully classified using the selected VIs, with the water-absorption band VIs performing better than the chlorophyll-absorption band VIs and other VIs. Similarly, for predicting the SGVS, the water absorption band indices (R² = 0.79 ± 0.04 and RMSE = 0.12 ± 0.01) outperformed the chlorophyll absorption band indices when using RF. Leave-one-out-location/year cross-validation revealed pronounced variation in model transferability driven by environmental and temporal domain shifts. RF consistently outperformed ANN, showing greater robustness to inter-site heterogeneity and interannual variability, whereas performance degraded most in spectrally distinct environments or atypical seasons. Interestingly, RDIS_3b (1280, 1250, 1180 nm), NDIS_2b (2190, 1510 nm), and NDWI2 (860, 1241 nm) were identified as the most critical predictors in the RF models, across merged and separated datasets. These findings demonstrate the potential of spectral signatures, particularly water-absorption band spectral indices, for quantitative phenotyping of stay-green as a proxy for drought tolerance in maize breeding programs; however, multisite, multiyear calibration is needed to enhance generalizability.
Accurate, field-scale mapping of crop growth stages is critical for supply-sensitive vegetable production, where timely harvests require detailed phenological information. Consecutive growth stages often involve rapid and subtle morphological changes and are influenced by challenging open-field conditions, which frequently result in misclassification when stages are treated as independent, discrete categories. To address this issue, CropMap is proposed as a growth-stage mapping framework that integrates Hierarchical Semantic Segmentation Networks (HSSN) with multispectral unmanned aerial vehicles (UAVs) imagery. CropMap incorporates the structured biological progression of crop development into the learning objective through tree-based label constraints, allowing the model to recognize phenological continuity and reduce confusion between adjacent stages. The framework is evaluated on the publicly available National Information Society Agency of Korea (NIA) field crop growth-stage dataset, a large-scale, multi-institutional UAV dataset containing 337,665 multispectral patches across six hierarchically related growth stages of Chinese cabbage and radish, curated by the NIA. CropMap achieves a test-set mean Intersection over Union (mIoU) of 0.5382, representing a 5% relative improvement over the best-performing transformer baseline (SegFormer; mIoU = 0.5124). Performance varies across classes: background separation is strong (IoU = 0.9128) and the rosette stage is well distinguished (IoU = 0.6354), while the leaf expansion stage remains the primary challenge (IoU = 0.3541), reflecting the inherent difficulty of mapping this spectrally and morphologically transitional class. These findings indicate that hierarchy-aware learning reduces inter-stage confusion for most phenological classes, but transitional growth stages remain a significant limitation for field-scale deployment. The framework provides a foundation for stage-resolved crop monitoring to support harvest timing and supply forecasting in high-value vegetable systems.
Plant diseases have continued to threaten agricultural productivity, while manual inspection methods have remained inefficient and prone to subjectivity. This study proposed and assessed a hybrid framework integrating a Convolutional Neural Network (CNN) with a Large Language Model (LLM) to perform image-based plant leaf disease classification accompanied by interpretable diagnostic explanations. EfficientNetV2-M was employed as the visual backbone and trained on 11 selected classes of apple, grape, and potato leaf images derived from the PlantVillage dataset. A structured data splitting strategy was applied to ensure reliable model validation and unbiased testing. The classification capability of the CNN component was examined through standard multi-class evaluation indicators, including class-wise predictive consistency and error distribution analysis. Experimental results indicated that the model delivered highly consistent predictions, reaching a peak test accuracy of 99.79%, reflecting its robustness in distinguishing visually similar disease patterns. To overcome the black-box limitation, prediction outputs were transformed into structured prompts and processed by GPT-4o to generate contextual explanations. The generated narratives systematically described observable symptoms, highlighted distinguishing characteristics, and suggested initial management actions. Overall, the proposed hybrid system demonstrated that combining high-performance visual recognition with language-based reasoning enhanced both diagnostic reliability and interpretability in digital agriculture applications.
Reproduction assets foundThe paper uses the public PlantVillage color dataset (11 apple/grape/potato classes, 9,385 images) and explicitly points to it in the DATA AVAILABILITY statement as the replication package data. The authors also provide a public Streamlit demonstration of their hybrid CNN–LLM system. No author analysis code or trained-Dataset · publicaper. The research was conducted for academic purposes, and no financial, commercial, or
personal relationships influenced the study design, data analysis, interpretation of results, or
preparation of the manuscript.
DATA AVAILABILITY
The data associated with this study are publicly available online in the replication package.
[https://github.com/spMohanty/PlantVillage-Dataset/tree/master/raw/color]
AUTHOR CONTRIBUTIONS
Frenky Riski Gilang Pratama: Conceptualization; Programming and coding
implementation; Methodology; Writing-Original Draft. Sugiarto Surono: Conceptualization;
Methodology; Supervision; Writing-Review & Editing. Aris Thobirin: Proofreading Paper;
Writing-Review & Editing; FunOpen asset ↗https://github.com/spMohanty/PlantVillage-Dataset/tree/master/raw/colorpdf-raw-page:10 lines:1-52Code / dataset availability confirmedCrossref · Europe PMC · checked 5 Sept 2026
Shoot apical meristem (SAM) homeostasis integrates environmental and genetic cues to regulate growth dynamics that drive biomass accumulation and crop yield; however, no robust, non-destructive, quantitative proxy has been established for modeling or monitoring SAM-homeostasis-associated dynamics. Here, we developed a novel robot-based 3D imaging system and a custom pot-chamber gas exchange system to non-destructively measure plant occupation volume (POV) and whole-plant photosynthetic rate in wild-type Arabidopsis plants and nine mutants with disrupted SAM homeostasis. We demonstrate that POV robustly captures 3D plant architecture, whereas whole-plant photosynthetic rate serves as a superior proxy for optimal growth dynamics and final biomass associated with SAM homeostasis, outperforming conventional traits such as leaf number, leaf size, total leaf area, and rosette diameter. The strong positive correlations among POV, whole plant photosynthesis, and biomass accumulation establish a powerful new framework for quantitative studies of SAM homeostasis and data-driven evaluation of plant architecture.
Reproduction assets foundThe paper's authors explicitly state that the Python source code for whole-plant leaf-area segmentation, 3D point cloud processing, POV calculation, and Mask3D-based segmentation is publicly available on GitHub at https://github.com/songqingfeng/AtPOVcalculator. This is a paper-specific, public, actionable analysis/PhDCode · publicThe Python source code for whole-plant leaf-area segmentation and calculation is publicly available on GitHub ( https://github.com/songqingfeng/AtPOVcalculator ).Open asset ↗songqingfeng/AtPOVcalculatorlines:224-233Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published1 Aug 2026International Journal of Innovative Science and Research TechnologyCited by 0 · OpenAlex ↗
This paper presents an IoT and edge-AI enabled autonomous agricultural robot that performs early plant disease diagnosis and precision irrigation on a single mobile platform. Unlike earlier automated farming systems that rely on visible-spectrum (RGB) imagery and simple threshold-based watering, the proposed system fuses RGB and nearinfrared (NIR) imagery to compute the Normalised Difference Vegetation Index (NDVI), enabling detection of physiological plant stress several days before visible lesions appear. Leaf images are classified using a lightweight attention-guided convolutional neural network that combines a MobileNetV3 backbone with a Convolutional Block Attention Module (CBAM), allowing the network to focus on lesion-relevant channels and spatial regions while remaining compact enough for real-time inference on an ESP32-S3 edge controller. Irrigation and pesticide-spray decisions are no longer governed by a rigid binary threshold; instead, a Mamdani-type fuzzy inference engine fuses soil moisture, ambient temperature, and the NDVI-derived stress index to compute a proportional, continuously variable actuation signal, reducing both water wastage and false triggering. The robot streams sensor readings, classification results, and actuation logs to a cloud dashboard over Wi-Fi/MQTT so that farmers can monitor crop health and irrigation status remotely and receive real-time alerts. Experimental evaluation on a prototype platform shows that the proposed attention-guided model improves disease-classification accuracy over a baseline CNN, the NDVI-assisted pipeline detects stress earlier than colour-only analysis, and the fuzzy irrigation controller reduces water consumption relative to the binary threshold scheme while maintaining optimal soil-moisture levels. The results indicate that combining multispectral sensing, attention-based lightweight deep learning, and fuzzy control on a single autonomous platform is a practical and scalable route towards sustainable, resource-efficient precision agriculture.
Image-based techniques are widely used in plant disease classification to support agricultural productivity and facilitate early detection. This study presents a comparative analysis between a baseline convolutional neural network (CNN) and a ResNet50-based transfer learning model for tomato leaf disease classification. A publicly available dataset containing five categories—tomato bacterial spot, tomato late blight, tomato septoria leaf spot, tomato yellow leaf curl virus, and healthy leaves—was used in the experiments. Model performance was evaluated using several standard classification metrics, with emphasis on overall accuracy. The baseline CNN achieved an accuracy of 97.0%, whereas the ResNet50 model reached 99.6%. The results demonstrate that the ResNet50 model produces more stable and reliable predictions, particularly when distinguishing between visually similar disease classes. These findings confirm that transfer learning can effectively improve classification performance in plant disease recognition tasks.
Early screening of tomato leaf diseases is important because foliar symptoms can reduce plant vigor and delay appropriate crop management. This study develops an Android-based tomato leaf disease classification system using MobileNetV2 as a lightweight convolutional neural network (CNN) architecture. The contribution of this work is the integration of model training, independent testing, and on-device Android deployment that supports camera and gallery inputs without relying on server-side computation. The dataset consisted of 1,200 balanced tomato leaf images from five classes: bacterial spot, late blight, target spot, tomato yellow leaf curl virus, and healthy leaf. Images were resized, normalized, augmented for training, and divided into training, validation, and independent testing subsets. The model obtained 94.12% training accuracy, 93.00% validation accuracy, and 89.00% independent test accuracy. The confusion matrix showed that tomato yellow leaf curl virus was classified without error, whereas bacterial spot, late blight, target spot, and healthy leaves produced several misclassifications because of similar lesion and discoloration patterns. The results show that MobileNetV2 is suitable for lightweight mobile disease screening, although larger field datasets, cross-validation, model comparison, and explainability analysis are still needed for broader deployment.
In-season fine-scale (i.e., within-field experiment plot scale) crop grain yield (GY) prediction is critical for optimizing inputs, minimizing environmental impacts, and supporting sustainable food production. Traditional approaches, such as field surveys, are often costly and inefficient over large areas. As an alternative, remote sensing combined with crop simulation models (CSMs) has been increasingly applied for in-season GY prediction. This study investigates the potential of integrating Uncrewed Aircraft Systems (UAS)-based remote sensing data, deep learning, and CSMs to predict maize and soybean GY using a data assimilation approach. UAS multispectral imagery was collected, along with field-measured maize above-ground biomass (AGB) and soybean leaf area index (LAI) during the 2022 and 2023 growing seasons at experimental fields in Brookings, South Dakota. Maize AGB was measured at two growth stages, while soybean LAI was collected across four stages. One-dimensional convolutional neural networks (1D-CNNs) were used to estimate maize AGB and soybean LAI from canopy spectral, textural, and structural features derived from UAS imagery. These UAS and deep learning–derived crop traits were assimilated into DSSAT-Maize and DSSAT-Soybean models to optimize parameters, and the optimized models were subsequently used to predict GY. For maize, the DSSAT-Maize model achieved an R² of 0.62, an RMSE of 717.8 kg ha⁻¹, and an rRMSE of 6.7% for GY prediction. For soybean, the DSSAT-Soybean model achieved an R² of 0.81, an RMSE of 207.3 kg ha⁻¹, and an rRMSE of 4.9%. Overall, these results highlight the potential of combining high-resolution UAS data and deep learning–derived crop traits within a CSM framework through data assimilation, enabling fine-scale, in-season yield predictions and supporting precise agricultural management.
Accurate and timely crop yield estimation is fundamental for global food security, agricultural policy, and farm management. The Copernicus Sentinel-2 constellation has catalyzed a paradigm shift in Earth observation for agriculture, enabling field and sub-field scale monitoring. This review synthesizes recent advances in crop yield estimation that leverage Sentinel-2 data. A dominant theme is the transition from regional-scale to high-resolution field-level assessments, driven by three approaches: (i) empirical models using vegetation indices coupled with machine and deep learning (e.g., Random Forest, Convolutional Neural Networks); (ii) integration of process-based crop growth models (e.g., WOFOST, SAFY) through data assimilation of Sentinel-2 derived biophysical variables such as Leaf Area Index; and (iii) data fusion of Sentinel-2 with Sentinel-1 Synthetic Aperture Radar to overcome cloud cover. The synthesis shows that Sentinel-2-based frameworks can explain a large fraction of within-field yield variability, while performance remains constrained by limited ground-truth data, cloud gaps, and model transferability. Looking ahead, knowledge-guided models, self-supervised foundation-model pre-training, lightweight edge workflows, improved ground observations, and multi-sensor fusion are key pathways toward robust, operational decision-support tools for precision agriculture.
Plant stress recognition plays a vital role in precision agriculture by enabling the early detection of diseases, insect infestations, and nutrient deficiencies that adversely affect crop productivity. Although deep learning models have achieved promising results, existing CNN-, Transformer-, and hybrid architectures often struggle to capture complex spatial dependencies, distinguish visually similar stress symptoms, and handle class imbalance in multi-label classification. To address these challenges, this paper proposes Q-TriLSTM-Vision, a novel hybrid deep learning framework that integrates an EfficientNet-B0 visual encoder, a tri-stream long short-term memory (TriLSTM) network, and a lightweight quantum-inspired interference gate. Unlike quantum computing-based approaches, the proposed interference mechanism is implemented entirely using classical neural operations to enhance feature representation without requiring quantum hardware. The model further employs an entanglement-inspired attention fusion module, focal binary cross-entropy loss with label smoothing, weighted sampling, and per-class threshold calibration to improve discriminative learning and minority-class recognition. The proposed framework was evaluated on the OLID-I dataset and further validated on the PlantVillage and PlantDoc benchmark datasets. Experimental results demonstrate that Q-TriLSTM-Vision achieved Macro-F1 scores of 0.9127, 0.9624, and 0.8975 on OLID-I, PlantVillage, and PlantDoc, respectively, outperforming representative CNN-, Transformer-, and hybrid deep learning models while achieving lower Hamming loss and improved recall. Cross-validation, ablation studies, and statistical significance analysis further confirm the robustness and effectiveness of the proposed framework. Overall, Q-TriLSTM-Vision provides an accurate, computationally efficient, and reliable solution for intelligent plant stress recognition in precision agriculture.
Reproduction assets foundThe paper's plant-phenotyping measurements are based on the public OLID-I plant stress image dataset hosted on Kaggle, which the authors explicitly identify as the data source for training and evaluating Q-TriLSTM-Vision. No author analysis code, trained models, or supplementary assets with explicit public availabilityDataset · publicusion method for incorporating the multi-directional representations. This combination is unprecedented in any other existing hybrid plant stress recognition model.
3
Materials and methods
3.1
Data collection
For this study, the data were gathered from the publicly available OLID-I plant stress image dataset hosted on Kaggle: ( https://www.kaggle.com/datasets/raiaone/olid-i ). We chose it mainly because it includes leaf images from multiple vegetable crops, and those images show both healthy and stressed states, too. In the implemented code, the dataset is pulled in through the KaggleHub package, while the folder structure is scanned in an automatic way. During this scan, the script tries toOpen asset ↗Kaggle · raiaone/olid-ilines:51-61Code / dataset availability confirmedCrossref · checked 14 Sept 2026
Sugarcane plant diseases pose a significant threat to agricultural productivity, yet early and accurate identification remains challenging for farmers due to the limitations of manual inspection. This study proposes a sugarcane leaf disease classification system using ConvNeXt V2 Tiny, a modern convolutional architecture with a Global Response Normalization (GRN) mechanism, combined with an ensemble Stratified K-Fold Cross Validation strategy (K=6) to improve generalization on real-world field data. A dataset of 2,948 leaf images spanning five classes (Red Rot, Mosaic, Rust, Yellow Leaf, and Healthy) was used, with field-collected images held out as a fixed test set. The ensemble model achieved a mean validation accuracy of 98.49% ± 0.58% across six folds and a test accuracy of 98.39% on 427 unseen field images, with macro-average precision, recall, and F1-score each reaching 98%. ConvNeXt V2 Tiny substantially outperformed ResNet-50 (87.35%) and EfficientNetV2-S (83.37%) under identical experimental settings, demonstrating superior generalization across the domain gap between curated and field data. The primary contribution of this study is the first application of ConvNeXt V2 Tiny with ensemble K-Fold strategy for sugarcane disease classification, offering high accuracy with moderate computational complexity (28.6M parameters) and practical deployability, as demonstrated through the SugarScan web application.
Reproduction assets foundThe paper's phenotyping inputs include a public Kaggle dataset (Sugarcane Leaf Disease Dataset, SLD) of sugarcane leaf disease images used for training/validation, plus field-collected images. Only the Kaggle dataset qualifies as a paper-specific public asset with an authors' URL; no author analysis code, trained modelDataset · publicsecondary data from the Sugarcane Leaf Disease Dataset (SLD)
available publicly on Kaggle (https://www.kaggle.com/datasets/pritpal2873/sug arcane-leaf-disease-dataset)Open asset ↗Kaggle · pritpal2873/sugpdf-page:2 lines:54-60Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published31 Jul 2026World Journal of Advanced Engineering Technology and SciencesCited by 0 · OpenAlex ↗
Proper diagnosis of plant disease and nutrient deficiency is crucial in enhancing crop productivity, reducing yield losses and early agricultural interventions. This paper introduces the machine learning approach for automated identification of plant diseases and their severity by analyzing the images of plant leaves. The proposed framework involves image preprocessing, feature extraction, classification and severity estimation, which would enable the accurate identification of various types of disease and estimation of their severity levels while detecting nutrient deficiencies. An extensive image dataset of healthy, diseased and nutrient deficient leaves was used to train and test the models. As illustrated by the experimental results, the proposed framework outperforms the existing machine learning and deep learning methods for plant disease identification and nutrient deficiency detection with the classification accuracy of 98.76% and 98.14% respectively. In addition, the severity assessment module estimates well to enable accurate pesticide and nutrient application, which minimizes chemical use. The proposed framework provides a scalable, efficient, and precise approach for smart crop health monitoring and precision agriculture applications.
Broccoli is a globally significant vegetable, but climate change and soil salinization increasingly threaten its productivity. Precise seedling phenotyping is essential for selecting salt-tolerant germplasm, yet traditional manual methods are labor-intensive and error-prone. This study develops LBD-PointNet++, an optimized 3D point cloud semantic segmentation model for automated phenotypic parameter extraction of broccoli seedlings at the germination and early developmental phase under salt stress. High-fidelity 3D point clouds were reconstructed from a precision three-view imaging system using Structure from Motion (SfM) algorithms. LBD-PointNet++ introduces three core optimizations: (1) a Large Kernel Attention (LKA) mechanism using 3D sparse decomposition to capture long-range global dependencies; (2) a Dual Uncertainty and Shape-Adaptive Sampling (DUSAS) mechanism to preserve high-frequency features of fragile stems and margins; and (3) a joint Boundary-Aware Nested Contrastive and Adaptive Varifocal Joint Loss (BNCV-Loss) to effectively isolate overlapping leaves. Experimental results demonstrate superior performance, achieving an overall mean Intersection over Union (mIoU) of 88.07% across all three categories (Leaf, Stem, and Pot) and a Mean F1-score of 93.48%. Compared to state-of-the-art Transformer architectures like PTv3, LBD-PointNet++ achieves higher accuracy with less than 6% of the parameter volume and over twofold faster inference speed. Furthermore, dynamic monitoring across NaCl gradients (0-250 mmol/L) revealed a potential non-linear threshold effect, identifying 100 mmol/L as a preliminary phenotypic threshold under these conditions. Beyond this threshold, growth inhibition intensified rapidly; At 250 mmol/L, plant height decreased by 54.43% and the 3D entity volume shrank to approximately one-fifth of the control group. In summary, LBD-PointNet++ provides a high-efficiency solution for phenotypic identification and digital breeding of salt-tolerant Brassicaceae crops.
Common beanGrapevineLeafClassificationObject detectionStress / disease detectionDisease symptoms / severity
Plant leaf disease is a grave risk to crop suitability and agricultural sustainability, and therefore, the ability to make early and accurate diagnosis is a mandatory need in the contemporary precision farming system.In recent years, deep learning has gained significant attention for image-based plant disease detection.Despite its effectiveness, model performance can be influenced by factors such as redundant feature representations and sensitivity to hyperparameter selection.To address these challenges, this study proposes a nested hybrid optimization framework that combines the Cuckoo Search Algorithm (CSA) for channel selection with the Beluga Whale Optimization Mechanism (BWOM) for tuning the hyperparameters of a Convolutional Neural Network (CNN).The proposed approach is independently evaluated on bean and grape leaf datasets under consistent experimental conditions to assess its disease classification performance.In addition to strong predictive performance, the framework incorporates explainable AI (XAI) techniques, namely Gradient-weighted Class Activation Mapping (Grad-CAM) and Gradientweighted Class Activation Mapping Plus Plus (Grad-CAM++), to enhance model interpretability.These approaches highlight the most significant visual features influencing predictions, thereby providing valuable insights for agronomists and fostering trust in AI-based systems.Experimental results show that the proposed CSA-BWOM optimized CNN achieves classification accuracies of 99.61% and 99.38% on the bean and grape datasets, respectively, outperforming baseline CNN models and exhibiting competitive performance when compared to a few existing approaches.
Identifying the complex leaf disease is a major challenge in crop production. Traditional Capsule Networks face challenges in capturing the complex features associated with various plant diseases, even though they can effectively retain spatial hierarchies. To overcome challenges, this research presents a novel Capsule Networks framework for the identification of leaf disease and its classification. The research proposes a Feedback-Enabled Capsule Routing and a Capsule-Level Attention mechanism. The Feedback-Enabled Capsule Routing method that refines classification outcomes also introduces top-down contextual signals, whereas the Capsule-Level Attention mechanism dynamically weighs capsule outputs based on their relevance. Incorporating this, we also integrated Primary Capsule Squashing and Dynamic Routing. All methods are grounded in formal mathematical definitions. The XAI techniques are integrated to visualize the model's decision-making process for transparency. Empirical evaluations demonstrate the model's effectiveness in achieving classification accuracies of 98.97% and 98.49% in the PlantVillage and the PLD datasets, respectively. This framework addresses the trust gap in AI-driven agricultural diagnostics, allowing for a transparent and reliable method of disease detection in agricultural applications.
Rapid, non-destructive phenotyping is vital for early nutrient deficiency detection in plant research and agriculture. Nitrogen (N) and magnesium (Mg) deficiency are hard to distinguish with instruments measuring only chlorophyll content since both deficiencies lead to loss of chlorophyll. This study evaluated the suitability of a commercially available dual-excitation fluorescence sensor (measuring chlorophyll (Chl) and epidermal UV absorbing compounds) for identifying nutrient deficiencies in maize seedlings. Both N and Mg deficiencies in maize seedlings caused a decrease in the Chl index and an increase in the flavonol (Flav) index, although the response of the Flav index was much weaker under Mg deficiency. This happened in spite of a much larger increase in sugar concentrations under Mg deficiency. Furthermore, the nitrogen balance index (NBI) detected N deficiency earlier in leaves developing under nutrient deficiency than in those present before treatment application. Spatiotemporal analysis revealed distinct patterns of Flav index increase: N deficiency caused a marked increase in upper (younger) leaves, whereas Mg deficiency initiated Flav index increases at the tips of lower (older) leaves. Utilizing green- and red-light excitation of chlorophyll to infer epidermal anthocyanins turned out to be not straightforward in nutrient-deficient maize leaves. Taken together, these findings reveal advantages and limitations of Chl fluorescence in diagnosing nutrient stress and underscore the importance of understanding spatial-temporal nutrient dynamics for accurate early detection. Leaf age should be considered for fertilization decisions based on the NBI. A strategy is suggested how anthocyanins can be detected without problems.
Agriculture plays a crucial role in the Indian economy. Early detection of plant diseases is very much essential to prevent crop loss and further spread of diseases. Most plants such as apple, tomato, cherry, grapes show visible symptoms of the disease on the leaf. These visible patterns can be identified to correctly predict the disease and take early actions to prevent it. This can be overcome by the use of machine learning and deep learning algorithms. Hence, we are proposing a method that which is detecting the disease of a tomato plant from their leaf images. Here the process is performed with the deep learning algorithms Convolutional Neural Network (CNN), and MobileNet which is a one of the transfer learning method of CNN. Once after training the dataset with the algorithms, the accuracy of algorithms is compared and the images are classified. And the precautions are also provided for the classified plant.
Penyakit Basal Stem Rot (BSR) merupakan penyakit penting pada tanaman kelapa sawit yang dapat menurunkan produktivitas. Analisis berbasis citra dapat mendukung deteksi penyakit, tetapi karakteristik distribusi warna perlu dipahami sebelum tahap klasifikasi. Penelitian ini bertujuan menganalisis distribusi warna citra daun kelapa sawit kelas Healthy dan BSR menggunakan Histogram pada ruang warna RGB dan HSV serta mengevaluasi kemampuan diskriminatif fitur warna. Dataset terdiri atas 2.438 citra dari repositori terbuka Roboflow. Setiap citra diproses melalui resize 224×224 piksel, Gaussian Blur, konversi RGB ke HSV, serta ekstraksi 12 parameter statistik berupa mean dan standar deviasi. Perbedaan distribusi antar kelas diuji menggunakan Mann–Whitney U dan besarnya perbedaan dihitung menggunakan Cohen's d. Kemampuan diskriminatif G_Mean, S_Mean, dan V_Mean dievaluasi menggunakan Receiver Operating Characteristic (ROC) dan Area Under the Curve (AUC). Hasil menunjukkan ketiga fitur memiliki perbedaan signifikan (p-value = 0,0000) dengan ukuran efek besar. Nilai AUC G_Mean, S_Mean, dan V_Mean masing-masing sebesar 0,9794, 0,7362, dan 0,9790. Hasil menunjukkan bahwa analisis distribusi warna dapat mengidentifikasi fitur diskriminatif sebagai dasar pemilihan fitur dan perancangan pra-pemrosesan citra BSR.
The prompt identification and precise categorization of chili plant diseases are crucial for promoting sustainable agriculture and reducing crop losses due to pests and pathogens. This research presents an automated detection method using the Single Shot Detector (SSD) framework, with ResNet-50 as the backbone and a Feature Pyramid Network (FPN) to manage multi-scale feature representations. Dataset including 14,248 images of chili plants, categorized into six classes like Leaf Spot, Rotten Fruit, Healthy Fruit, Healthy Leaf, Mosaic Curl, and Yellowing, underwent preprocessing involving segmentation, resizing, and augmentation, followed by a division into 90% training data and 10% testing data. Transfer learning was implemented using a COCO-pretrained SSD ResNet-50 FPN model, enhanced with cosine-decay learning-rate scheduling and momentum optimization. The assessment results indicated an overall accuracy of 92.7%, with the highest F1-scores achieved for Healthy Fruit (0.965) and Rotten Fruit (0.967). Under the COCO evaluation protocol, the model achieved an mAP@0.5 of 91.5% and mAP@[0.5:0.95] of 65.2%. Model ran at approximately 30 FPS on an NVIDIA T4 GPU. Reduced precision values were noted for Leaf Spot (0.866) and Mosaic Curl (0.850), suggesting a propensity for misclassification due to visual similarities among disease symptoms. Nonetheless, all classes attained F1-scores exceeding 0.86, illustrating the robustness of the proposed model. Importantly, the SSD-ResNet-50 approach offers both efficiency and accuracy within a single pipeline, enabling rapid inference practical for real-world applications. These findings emphasize the potential of deep learning-based solutions to strengthen plant disease monitoring systems. In conclusion, SSD with ResNet-50 FPN provides an effective and scalable methodology for the automated identification of chili plant diseases, contributing directly to sustainable agriculture and improved crop management practices.
Camera-based visual sensing provides a non-destructive and scalable approach for monitoring strawberry diseases and pests in greenhouse environments. However, greenhouse images acquired under practical cultivation conditions often contain early-stage tiny lesions, complex leaf backgrounds, uneven target scales, illumination variations, and partial occlusions, making accurate and efficient visual detection challenging. To address these issues, this study proposes YOLOv8n-DSLW (YOLOv8n enhanced by Dense reuse, Shuffle attention, LSKA-LAMP lightweight modeling, and Wise-IoU optimization), an AI-enabled vision-sensing detection model based on YOLOv8n for tiny strawberry disease and pest detection. Specifically, Shrink Residual Dense Block (ShrinkRDB) dense connection blocks and the C2f with Shuffle Attention (C2fSA) module are introduced to preserve weak lesion textures and suppress background interference in greenhouse visual data. A high-resolution P2 detection layer combined with Wise-IoU (WioU) dynamic regression loss is further incorporated to enhance tiny-target perception and localization. In addition, the Spatial Pyramid Pooling-Fast with Large Separable Kernel Attention (SPPF-LSKA) module strengthens contextual modeling under occlusion and clutter, while Layer-Adaptive Magnitude-based Pruning (LAMP) is adopted to mitigate model redundancy and improve the accuracy-efficiency balance. Experiments on a self-collected greenhouse strawberry disease and pest dataset show that YOLOv8n-DSLW achieves a mean Average Precision at 0.5 IoU threshold (mAP@0.5) of 94.3% and a mAP@0.5:0.95 of 77.5%, outperforming the YOLOv8n baseline. The final model has a parameter count of 4.386 M and a computational cost of 27.6 GFLOPs, achieving a frame rate of 45 FPS on the test workstation. It shows application potential for real-time visual monitoring in greenhouses under controlled data acquisition conditions. The results demonstrate that the proposed method improves tiny lesion detection under dense targets, complex backgrounds, and leaf occlusions, providing an AI-enabled vision-sensing framework for automated strawberry health monitoring in greenhouses. Nevertheless, due to limitations associated with imaging equipment, dataset representativeness, and the inherent constraints of the algorithm, further optimization and validation are required to support large-scale field deployment.
Reproduction assets foundThe paper's self-collected greenhouse strawberry disease/pest image dataset (with COCO annotations and train/test splits) is explicitly stated as publicly deposited on GitHub at the allowed URL. No author analysis code or trained model checkpoint is mentioned as publicly available.Dataset · publicThe dataset used in this study, including the training and independent test subsets, has been uploaded to a GitHub repository for dataset verification and is available at: https://github.com/dataset-review-2026/strawberry-dataset (accessed on 26 July 2026).Open asset ↗dataset-review-2026/strawberry-datasetlines:111-131Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Published30 Jul 2026International Journal of Advanced Research in Science Communication and TechnologyCited by 0 · OpenAlex ↗
Agriculture continues to be one of the principal contributors to the economy and food security of developing nations, yet farmers regularly face difficulties such as unpredictable weather, variable soil conditions, and crop diseases that reduce quality and income. This paper presents a Crop Prediction and Plant Disease Detection System that integrates machine learning and deep learning techniques within a single web-based platform. The system accepts agricultural parameters — nitrogen, phosphorus, potassium, temperature, humidity, pH, and rainfall — and applies a Random Forest regression model to estimate the expected crop. In parallel, it allows farmers to upload images of crop leaves, which are pre-processed and classified by a Convolutional Neural Network (CNN) to identify plant diseases and recommend suitable treatment. The application is built using ASP.NET Core for the user-facing interface, authentication, and dashboard, while a Python-based REST API hosts the machine learning and deep learning models; Microsoft SQL Server is used for persistent storage of user data, predictions, and disease records. The proposed system combines two traditionally separate functions — crop estimation and disease diagnosis — together with treatment recommendations and prediction history, into a single decision-support tool for precision agriculture, and was validated through unit, integration, and system-level testing
Introduction Timely and accurate detection of plant diseases is essential for ensuring global food security and supporting sustainable agriculture. Conventional diagnostic approaches, such as manual inspection and laboratory testing, are often time-consuming, labor-intensive, and impractical for large-scale or remote agricultural environments. Although deep learning models, particularly Convolutional Neural Networks (CNNs), have significantly improved automated plant disease classification, they often lack interpretability and struggle to generalize under diverse field conditions. Methods This study proposes AgriX-SENet, an explainable deep learning framework that integrates Squeeze-and-Excitation (SE) blocks with a DenseNet121 backbone to enhance disease classification performance. The SE blocks recalibrate channel-wise feature responses to emphasize disease-relevant information while suppressing background noise. To improve model transparency, Grad-CAM, SHAP, and LIME were incorporated to provide visual and feature-level explanations of the model’s predictions. The framework was trained and evaluated using the Plant Pathology 2020 dataset containing four classes: healthy, rust, scab, and multiple diseases. Results AgriX-SENet achieved a training accuracy of 97.47% and a validation accuracy of 95.07%, outperforming fourteen state-of-the-art deep learning models. The classification report demonstrated high precision and recall across most disease categories, although the scab class exhibited comparatively lower recall, indicating an opportunity for further improvement. The explainability analyses consistently showed that the model focused on pathologically relevant regions of leaf images, validating the reliability of its predictions. Discussion The proposed AgriX-SENet framework effectively combines high classification performance with model interpretability, addressing a key limitation of existing CNN-based plant disease detection systems. Its ability to provide accurate and explainable predictions makes it a promising solution for scalable agricultural diagnostics. Future work will focus on improving classification performance for challenging disease categories and optimizing the framework for deployment on mobile and edge computing devices to enable real-time field applications.
Reproduction assets foundThe paper trains and evaluates AgriX-SENet on the public Plant Pathology 2020 (FGVC7) Kaggle image dataset, which is the paper-specific plant image input for its disease-classification measurements. No author analysis code, trained model checkpoints, or supplementary code/data deposit is mentioned; the data statement (Dataset · publicPlant Pathology 2020 - Fgvc7 . Available online at: https://www.kaggle.com/competitions/plant-pathology-2020-fgvc7/data .Open asset ↗Kaggle · plant-pathology-2020-fgvc7lines:895-974Plant phenotyping relevance matchOpenAlex · checked 16 Sept 2026
The present study validates a custom multi-sensor system for high-resolution, non-phenotyping. The photogrammetric workflow was optimised by evaluating image density, algorithms, and camera calibration. Beyond validation, a case study demonstrated the system’s capacity to monitor early plant development following biostimulant treatment. Technical assessment revealed that prior internal camera calibration was unnecessary for the optics used. A reduced dataset of 120 images yielded reconstruction accuracies statistically comparable to full 360-image sets ( p > 0.05), confirming potential for maximised throughput efficiency by reducing processing time by approximately 66% without compromising data integrity. Analysis demonstrated that algorithmic settings determined reconstruction accuracy. Active parameter tweaks were essential for maximising surface area precision across all plant species ( R 2 ≥ 0.98). Conversely, volumetric accuracy required deactivation of these tweaks to maintain mesh consistency. While surface area estimation remained robust, volumetric precision showed species-specific variability driven by morphological complexity. Baseline parameters for accurate plant health scaling were established by defining species-specific vegetation index ranges (e.g., 0.38–0.82 for C. sativus ). The platform’s robustness was validated through a longitudinal study evaluating four treatments: yeast autolysate (A), a fungal biostimulant (F), their combination (AF), and a control (C). The system captured distinct morpho-physiological responses, demonstrating that autolysate-based treatments (A and AF) significantly enhanced biomass growth. The developed multi-sensor system recorded surface area expansions of 122% and 102% relative to the control ( p < 0.001), alongside a 110% increase in biological height. Fidelity of these 3D reconstructions was substantiated by a strong correlation ( R 2 = 0.96) between the 3D-derived leaf area index and ground-truth measurements. A key innovation of the pipeline is the integration of vertical distribution metrics as descriptive statistical tools, enabling high-resolution characterisation of canopy architecture. The plant health status metric evidenced enhanced physiological resilience in variants A and AF. Gravimetric analysis corroborated the non-destructive findings, confirming significant increases ( p < 0.001) in dried shoot weight of 77% (A) and 79% (AF). The validated system decoupled structural biomass from physiological health, offering broad utility across diverse phenotyping tasks. Such functionality streamlines the valorisation of industrial by-products into biopreparations, driving progress in sustainable agriculture.
Introduction Wheat leaf biomass is a key indicator of crop growth, nitrogen status, and yield potential, and its accurate estimation is essential for precision agriculture. Unmanned aerial vehicle (UAV) remote sensing provides multi-stage phenological observations for non-destructive biomass monitoring. However, existing approaches often fail to capture the superimposed temporal patterns inherent to crop phenology, including short-term physiological fluctuations driven by management events and long-term seasonal growth trends, as well as the cumulative causal effects of early-stage conditions on final biomass accumulation. Methods This study proposed a dual-branch perception and hybrid attention integrated framework (DBAFN) for temporal estimation of wheat leaf biomass from UAV multi-temporal observations across key growth stages. Results and discussion Experimental results demonstrated that the DBAFN achieved the best performance, with the coefficient of determination (R²) of 0.87, root mean square error (RMSE) of 38.41 g/m², mean absolute error (MAE) of 27.77 g/m², and relative RMSE (RRMSE) of 17.29%. Overall, the proposed framework provided an effective solution for temporal biomass estimation and demonstrated strong generalization capability, as further validated by independent experiments across different ecological regions and wheat genotypes (R² = 0.816-0.820). Compared with conventional machine learning models, the DBAFN showed consistently higher accuracy and lower prediction error. Multi-source feature analysis indicated that the combination of reflectance, vegetation indices, and canopy height provides the most accurate estimation. Ablation experiments further confirmed the effectiveness of each module in improving model performance. The SHapley Additive exPlanations (SHAP) analysis revealed that the canopy height and key spectral features contribute most to biomass prediction, highlighting the importance of integrating structural and physiological information. This study demonstrates that integrating multi-scale temporal dynamics, hybrid attention mechanisms, and transformer-based dependency modeling significantly improves the reliability of UAV-based biomass estimation. It offers a practical, data-driven pathway for intelligent crop monitoring and precision nitrogen management.
The ready-to-eat lettuce industry is rapidly expanding, increasing the need for reliable, scalable methods to assess seed germination and early growth under realistic soil conditions. This study presents an automated imaging-based approach for quantifying germination dynamics and seedling vigor using a low-cost multi-camera system under greenhouse conditions. Lettuce seeds were grown in soil either inoculated or non-inoculated with the soil-borne pathogen Rhizoctonia solani. Top-view images were acquired using commercial surveillance cameras and processed through a calibrated pipeline including geometric correction, color normalization, vegetation segmentation, clustering, and temporal tracking of emergence events. Seedling vigor was quantified through projected leaf area estimation. The proposed method enables accurate estimation of germination kinetics and growth dynamics under field-like conditions. Automated counts were validated against manual measurements at both intermediate and final time points, achieving high agreement in both cases. At the final assessment, the method reached R² = 0.98 and RMSE = 1.12, while at the midterm evaluation it achieved improved performance with R² = 0.998 and RMSE = 0.5, reflecting the lower complexity of plant structure at earlier growth stages. Results showed that pathogen inoculation significantly reduced both germination rate and seedling vigor, with up to 70% reduction in biomass accumulation. The proposed framework provides a robust, low-cost solution for high-throughput phenotyping of early plant development in soil-based systems, supporting scalable agricultural experimentation.
Abstract Deep learning has achieved remarkable success in rice disease diagnosis; however, existing methods often suffer from limited interpretability and poor robustness against open-world environmental noise. To address these challenges, this study proposes the Knowledge-Guided Multi-Task Rice Network (MTRNet) built upon a ResNet-50 backbone. Unlike conventional "black-box" models, MTRNet employs expert knowledge injection via a phytopathological matrix to explicitly disentangle disease features into Shape, Color, and Location attributes within a multi-head architecture. Furthermore, to mitigate false positives in complex field scenarios, a non-parametric Cascade Inference System (CIS)—comprising a biological grayscale filter and a visual consistency check—is introduced for robust Out-of-Distribution (OOD) detection and anomaly rejection. Experiments on a benchmark dataset of 5,932 field images, which primarily comprises four main rice diseases (Rice Leaf Blast, Brown Spot, Bacterial Leaf Blight, and Tungro), demonstrate that MTRNet achieves a diagnostic accuracy of 99.83%. Crucially, in an open-world robustness evaluation involving 1,000 non-agricultural noise samples, the proposed system achieved an 81.80% OOD rejection rate. By balancing diagnostic accuracy with structural transparency, this framework effectively narrows the gap between laboratory benchmarks and real-world agricultural applications.
Plant phenotyping plays a critical role in understanding plant health and improving agricultural productivity by enabling quantitative analysis of disease-related physiological characteristics.Among these, leaf diseases significantly impact crop yield and quality, necessitating accurate and automated phenotyping approaches.Traditional phenotyping methods rely on manual inspection or handcrafted feature extraction, which are time-consuming, prone to human error, and lack scalability under diverse environmental conditions.This study introduces a deep learning (DL)-based approach for image-based plant phenotyping, focusing on the classification of disease-affected traits.The developed method integrates segmentation-based region extraction, dual-branch feature learning, and attention-based feature fusion.Initially, input images are preprocessed and passed through a TransUNet-based segmentation module to isolate phenotypically relevant leaf regions while suppressing background interference.Both the original image and the segmented region are then processed using a RegNet-based feature extraction network to capture global structural information and localized disease-specific characteristics.The extracted features are fused using an attention-based mechanism, followed by fully connected layers for multiclass classification.Experimental results obtained on the controlled PlantVillage grape leaf dataset, which serves as a standardized benchmark for plant disease classification, demonstrate an overall classification accuracy of 97.8%, with precision, recall, and F1-score values of 97.7%, 97.9%, and 97.8%, respectively.In addition, the segmentation module achieves an Intersection over Union (IoU) of 94.1% and a Dice score of 96.8%, confirming its effectiveness in isolating relevant phenotypic regions.
Sebastian Tonn · Mon-Ray Shao · Jos de Wit · Rami Mousa · Iñigo Bañales · Roy A M van Beekveld · Arjen N Bader · Henriëtte D L M van Eekelen · Ric C H de Vos · Eefjan Breukink · Guido Van den Ackerveken
Abstract Effective plant disease phenotyping is crucial for resistance breeding, but traditional visual assessment is often inaccurate and inefficient. This is particularly challenging when breeding lettuce (Lactuca sativa) for resistance to downy mildew, given the obligate biotrophic lifestyle of the causal pathogen Bremia lactucae. We discovered that B. lactucae-infected lettuce exhibits patches of increased blue-green fluorescence (BGF) under UV-A excitation from 6 d post-inoculation, preceding visible symptoms. Co-localization of BGF with hyphae, visualized with trypan blue, indicates that BGF is induced by downy mildew colonization. We therefore investigated its potential for non-invasive disease detection and quantification, as well as the underlying physiological changes. Using a custom imaging system, we demonstrate that BGF leaf area correlates with downy mildew severity and can be automatically quantified via a U-Net-based convolutional neural network, enabling early, objective disease assessment. Exploring transcriptomic and metabolomic changes associated with BGF, we found that induction of the phenylpropanoid pathway led to accumulation of caffeoylquinic acids, whose fluorescence spectra overlap with that of BGF tissue, supporting the hypothesis that these compounds contribute to the fluorescence signal. BGF imaging offers a powerful tool for phenotyping in lettuce breeding and for identifying quantitative resistance traits that support durable downy mildew resistance.
Agriculture plays a pivotal role in ensuring global food security, economic stability, and sustainable development. Plant diseases significantly reduce agricultural productivity, resulting in substantial economic losses and threatening food supply worldwide. Early and accurate identification of plant leaf diseases enables timely intervention, minimizes crop damage, and enhances agricultural yield. Traditional disease diagnosis relies heavily on visual inspection by agricultural experts, making the process labor-intensive, subjective, and unsuitable for large-scale deployment. Recent advances in artificial intelligence, particularly deep learning, have transformed plant disease diagnosis by enabling automatic feature extraction and highly accurate image-based classification. This review presents a comprehensive analysis of recent developments in deep learning techniques for plant leaf disease identification. Various convolutional neural network (CNN) architectures, including AlexNet, VGGNet, ResNet, DenseNet, EfficientNet, MobileNet, Inception, and Xception, are critically reviewed along with modern transformer-based models such as Vision Transformer (ViT), Swin Transformer, and hybrid CNN–Transformer frameworks. The paper also examines transfer learning strategies, object detection methods including YOLO and Faster R-CNN, and semantic segmentation approaches such as U-Net and DeepLabV3+. Publicly available benchmark datasets, including PlantVillage, PlantDoc, AI Challenger, Cassava Leaf Disease, and Rice Leaf Disease datasets, are discussed in terms of dataset diversity, annotation quality, and practical applicability. Furthermore, image preprocessing techniques, data augmentation methods, evaluation metrics, and deployment considerations for mobile and edge devices are comprehensively reviewed. The paper identifies current research challenges, including dataset imbalance, environmental variability, model interpretability, computational complexity, and limited real-world generalization. Finally, emerging research directions such as explainable artificial intelligence, federated learning, multimodal learning, self-supervised learning, lightweight architectures, and edge AI are discussed to provide future research opportunities. This review serves as a valuable resource for researchers, practitioners, and agricultural technologists interested in developing robust, scalable, and intelligent plant disease identification systems.
Enhancing photosynthesis is an important approach to improve crop yields. Photosynthesis, as a key factor determining crop yield, is an important approach to increasing crop production and addressing global food security issues. Improving its efficiency is crucial in this regard. However, traditional photosynthetic phenotyping has long been a bottleneck in crop breeding due to time-consuming data collection. In this study, we simultaneously measured the spectral reflectance and the net photosynthetic rate (Pn) of soybean leaves to develop a high-precision model for estimating Pn based on hyperspectral data. By applying this model, we evaluated Pn in 219 soybean materials. A multi-environment genome-wide association study (GWAS) based on multi-environmental prediction Pn was carried out using the 3VmrMLM method, and 24 significant quantitative trait loci (QTLs) and four suggestive QTLs were identified. Among them, 24 QTLs overlapped with multiple previously reported QTL related to photosynthesis, chlorophyll content, quality, etc., or with genes related to key agronomic traits such as yield. Additionally, four new QTLs were discovered, and four candidate genes potentially associated with Pn were identified. Further, haplotype analysis identified their optimal haplotypes. This study presents a robust and nondestructive hyperspectral model for estimating the photosynthetic rate in soybeans, which is successfully applied to genetic analysis, yielding stable and biologically meaningful results. The approach offers an effective means to explore the genetic basis of photosynthesis and provides a solid theoretical foundation for large-scale, monitoring of soybean photosynthetic physiology.
Accurate and quick detection of plant leaf diseases is essential for precision agriculture to intervene promptly and boost crop yields. A new deep learning model called ResVNet has been introduced in this study. It combines the powerful local feature detection of ResNet152 with the global attention capabilities of Vision Transformer (ViT) and utilises Low-Rank Adaptation (LoRA) to accelerate fine-tuning. The PlantVillage dataset, which contains both healthy and diseased tomato samples, was used to train and test ResVNet. Experimental evaluation on the PlantVillage tomato dataset using stratified 5-fold cross-validation demonstrates that the proposed ResVNet model achieves a mean classification accuracy of 97.45%, along with superior macro-precision, macro-recall, and macro-F1 scores compared to existing deep learning architectures. The results of the confusion matrix and the ROC analysis validate its discriminatory power. The results highlight the potential of architectures strengthened with transformers in agricultural diagnostics. For real-time disease detection in the field, ResVNet is perfect for edge device deployment on drones and smartphones thanks to its high accuracy and adaptability. The application of Explainable AI (XAI) technologies for interpretability, integration with the Internet of Things (IoT), and multi-crop classification will all be explored in future studies. We will also look into model compression approaches so we can deploy efficiently in low-resource settings without sacrificing performance.
Introduction Banana production is increasingly threatened by fungal diseases such as Fusarium wilt and Black Sigatoka, posing severe risks to food security and agricultural economies. Recent image-based approaches using deep learning have shown high predictive capacity for plant disease recognition; however, their limited transparency, calibration uncertainty, and sensitivity to domain shifts can restrict their use in decision-support workflows that require auditability. Methods This study proposes an interpretable and calibrated Artificial Intelligence framework for multiclass banana disease-pattern characterization based on radiomic feature analysis of RGB leaf images. Radiomic features were extracted from HSV-segmented banana leaf regions, resulting in a dataset of 14,763 samples characterized by 103 quantitative descriptors and labeled as Healthy, Sigatoka, or Fusarium wilt race 1. Results Among the evaluated radiomics classifiers, the calibrated Random Forest achieved accuracy = 0.85, balanced accuracy = 0.84, macro-F1 = 0.84, and macro ROC-AUC OvR = 0.95 on the held-out test set. Bootstrap analysis yielded 95% confidence intervals of [0.8406, 0.8691] for accuracy and [0.8306, 0.8606] for balanced accuracy. Three deep learning baselines trained on the same partition achieved higher predictive performance: MobileNetV3 with accuracy = 0.96, macro-F1 = 0.95, and macro ROC-AUC OvR = 0.97; EfficientNet with accuracy = 0.96, macro-F1 = 0.97, and macro ROC-AUC OvR = 0.96; and ResNet-18 with accuracy = 0.96, macro-F1 = 0.96, and macro ROC-AUC OvR = 0.96. Discussion The CNNs produced strong classification performance on the evaluated repositories, and the radiomics approach demonstrated to be a complementary interpretable and explainable calibrated reference model. SHAP, LIME, permutation importance, accumulated local effects, calibration curves, and Brier score decomposition supported feature-level inspection of the final model.
Leaf disease diagnosis needs models that are accurate enough for agronomic use yet small enough for constrained computing settings. This study examines a late-attention MobileNetV2 design in which one Convolutional Block Attention Module (CBAM) is inserted between the last MobileNetV2 convolutional map and global average pooling. The experiments use 54,306 controlled-background PlantVillage images spanning 38 classes. Under a uniform saved-model re-evaluation, MobileNetV2 + CBAM obtained 97.17% accuracy and 97.15% weighted F1-score, whereas MobileNetV2 obtained 96.78% and 96.73%. On the converted models, paired testing gave a 0.64-percentage-point accuracy advantage for the CBAM variant (95% CI: 0.31–0.96; exact McNemar p
Plants encounter multiple abiotic stresses. Among them, heat and drought stress play a substantial role in reducing the agricultural productivity of commercial plants. Hence, wild and underutilized plants can be a potential alternative as they are naturally tolerant to extreme climatic conditions and are a rich source of nutrition. Manual stress and disease detection is a laborious and expensive process, and hence automation in this field is required to reduce agricultural losses. This study evaluates the prediction and detection of abiotic stress in Acacia senegal bipinnate leaves, exploring various stress-induced changes using machine learning (ML) algorithms and biochemical analysis. A. senegal , an underutilized edible desert legume, was grown under controlled greenhouse conditions. After 2 months, these plants were segregated into groups and subjected to heat and drought treatments. Image acquisition was performed to obtain a dataset of 3,454 images of A. senegal leaves. Physiological parameters, such as fresh and dry leaf weight, shoot length, number of leaves, and biochemical assays like antioxidant assay (DPPH), total phenolic content (TPC), and total flavonoid content (TFC), were determined. LC-MS/MS analysis was conducted to identify over 50 phytochemical compounds. A hybrid model was developed consisting of a fine-tuned EfficientNet-based Convolutional Neural Network (CNN) followed by a Support Vector Machine (SVM) for the binary classification of A. senegal leaves. The model distinguishes between healthy and stress-affected unhealthy leaves and achieved an accuracy score of 86.6%. This report provides a significant lead toward stress phenotyping and prediction of a bipinnate leaf plant using ML algorithms. The overall study is useful to understand how the stress encountered by arid plants alters the nutritional quality.
Reproduction assets foundThe paper's data availability statement explicitly makes the 3,454-image A. senegal leaf imaging dataset public on Zenodo and the ML implementation source code public on GitHub; both are paper-specific, public, and actionable.Dataset · publicThe plant leaf imaging data used in the work is publicly available at https://doi.org/10.5281/zenodo.16531486.Open asset ↗zenodo · 10.5281/zenodo.16531486html-lines:480-497Code · publicThe source code of the implementation is available at https://github.com/softwareinnovationslabBITS/CDRF_ASenegal_MLImagingOpen asset ↗github · softwareinnovationslabBITS/CDRF_ASenegal_MLImaginghtml-lines:480-497Plant phenotyping relevance matchEurope PMC · checked 14 Sept 2026
Rice leaf diseases pose a major threat to crop productivity and global food security. Therefore, rapid and accurate disease diagnosis is essential for effective crop management. Despite the advancements made by deep learning algorithms in automated disease detection, the majority of current deep learning-based models are built and evaluated on single-source datasets. Therefore, the generalization capability of these models is still questionable. In order to overcome this problem, this paper presents an interpretable deep learning approach for classifying diseases in rice leaves using ConvNeXt-Tiny and Grad-CAM. The proposed semantic class normalization technique is used to harmonize the heterogeneous data classes to form a seven-class classification system. The model performance was analyzed based on two available public datasets for rice leaf disease and a combination of both. After performing five individual runs, the average classification accuracies were calculated to be 99.13±0.37%, 95.28±0.63%, and 97.27±0.41% for Datasets 1, 2, and the combined dataset, respectively. The confusion matrix analysis indicated minor misclassification errors in the form of false positives for the disease classes which have similarities in appearance. The training and validation curves showed consistent learning with minimum overfitting. Moreover, Grad-CAM analysis confirmed the focus of the model on the disease-specific regions. These results collectively show that the combination of transfer learning, semantic class harmonization, and explainable AI is a stable platform to conduct automated rice disease detection. It is evident from these outcomes that reliable performance can be achieved even under heterogeneous imaging conditions. Consequently, there exists great potential for this approach in agricultural decision-support systems.
Plant disease detection is critical for sustainable agriculture and food security. While deep learning models achieve high accuracy in leaf disease classification, their black box nature poses limitations for trust and adoption among agricultural practitioners. This study presents a comparative evaluation of three convolutional neural network architectures (ConvNeXt-Tiny, MobileNetV2, and VGG16) for classifying potato, maize, and pepper leaf diseases, with emphasis on explainability through Gradient-weighted Class Activation Mapping (Grad-CAM). The experimental results demonstrate that ConvNeXt-Tiny achieves 99-100% accuracy across all plant species, MobileNetV2 attains 97-100% accuracy with lower computational requirements, and VGG16 yields 97-99.5% accuracy. Grad-CAM visualizations reveal that modern architectures precisely focus on lesion regions, whereas older models occasionally attend to irrelevant features such as leaf veins and edges. Misclassification analysis identifies shadows and natural leaf patterns as primary error sources. This research demonstrates that explainable artificial intelligence is not merely complementary but essential for developing trustworthy agricultural decision support systems.
Plant diseases have long been considered a major threat to global food production systems. Therefore, early diagnosis is vital to mitigate the risk of these diseases. This task can be challenging, as the number of harmful diseases is substantial. One technology that has gained widespread interest is artificial intelligence, specifically deep learning, which is used to identify plant diseases using leaf patterns. This paper presents two deep learning models, a custom CNN model and a transfer learning model based on the DenseNet-121 architecture. Experiments were carried out using the PlantCity dataset, which consists of twelve subsets of diverse crop species representing fruits, vegetables, and grains with variations among the subsets, including the number of classes, the subset sizes, class distribution, and visual complexity of disease symptoms. The two models were evaluated using multiple metrics, including accuracy, loss, precision, recall, and F1-score. Explainable AI using the LIME technique was deployed to better interpret the acquired results. Results showed that the developed transfer learning model based on DenseNet-121 had superior performance over the CNN model, with accuracies ranging from 86% to 99% across eleven experimented crops. In order to perform an independent experimental validation for the developed model, future work will focus on constructing a local crop dataset captured from Iraqi fields to evaluate the developed models based on the local environment.
Accurate assessment of plant nitrate status is critical for growth and productivity, yet early and non-destructive quantification remains challenging. Although Raman spectroscopy has been used to detect nitrate deficiency in plants, quantitative estimation of nitrate concentration from Raman spectra has not been demonstrated. Here, we evaluated whether Raman spectroscopy can be used to quantitatively predict leaf nitrate concentrations during early nitrogen stress. Two-week-old Pak Choi (C3) and Amaranthus (C4) plants were subjected to nitrate deprivation for 1–3 days, and Raman spectra were collected and correlated with nitrate concentrations determined by biochemical assays. A strong linear relationship was observed between nitrate concentration and the intensity ratio of the nitrate-associated Raman peak at 1046 cm - ¹ to the neighboring 1067 cm - ¹ peak. This relationship was consistent among plants of the same species and across different levels of nitrogen deficiency. Linear regression models achieved root-mean-square errors of 101 µg g - ¹ fresh weight (FW) in Pak Choi (~7% of nitrate under sufficient nitrogen) and 32 µg g - ¹ FW in Amaranthus (~19%), closely matching biochemical measurements and revealing species-specific nitrate dynamics. These findings demonstrate that Raman spectroscopy enables rapid, non-destructive, and quantitatively reliable estimation of leaf nitrate levels during early nitrogen stress, providing a promising platform for precision nutrient management and real-time plant phenotyping.
Early and accurate identification of plant diseases is essential for improving crop productivity and ensuring food security. Many existing deep learning-based plant disease classification methods rely solely on leaf images collected from a controlled environment, which limits their applicability in real-world agricultural conditions where symptoms may be visually unclear and influenced by environmental factors. To address these challenges, this study discusses AgriFusionNet, a context-aware multimodal deep learning framework that integrates leaf images, textual symptom descriptions, and environmental data for robust plant disease classification. The proposed architecture employs EfficientNet-B0 for visual feature extraction, BERT for semantic representation of symptom descriptions, and a lightweight multilayer perceptron for modeling environmental factors such as temperature, humidity, rainfall, and soil moisture. Features from all three modalities are fused into a unified representation to train the CNN model. The model is trained and tested upon the Context-Aware Multimodal Augmented PlantVillage dataset covering 38 plant diseases and healthy classes. Experimental results show that AgriFusionNet gives an overall accuracy of 98.94% on the dataset Context-Aware Multimodal Augmented PlantVillage, with competitive precision and recall and F1-score. The multimodal framework facilitates the co-learning of visual, semantic, and contextual environmental representations and the analyses of the confusion matrix and feature interactions give insights into cross-modal relationships. The proposed approach aims to explore context-aware multimodal representation learning for agricultural AI applications, with emphasis on integrating complementary visual, semantic, and contextual information.
Reproduction assets foundThe paper's data availability statement points to the Context-Aware Multimodal Augmented PlantVillage dataset (leaf images, symptom text, environmental data used for the phenotyping/classification analysis) deposited publicly on IEEE Dataport with a DOI matching an allowed URL.Dataset · publicPublicly available datasets were analyzed in this study. This data can be found here: Dataset. IEEE Dataport. https://dx.doi.org/10.21227/9jat-r836 [Accessed on August 2025].Open asset ↗IEEE Dataport · 10.21227/9jat-r836lines:1029-1047Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Qing Chang · Lixin Wang · Mallory L. Barnes · Darren L. Ficklin · Michael C. Benson · Kimberly A. Novick
Field / plotLeafWhole plant / canopy / plot / fieldObject detectionPhysiological trait estimationGrowth / time-series analysisLeaf traitsWater status / transpiration
Abstract Understanding the vulnerability of plants to more severe and frequent drought events and developing adaptive management strategies requires robust methods for quantifying long‐term changes in plant water stress (PWS). Most data‐driven explorations of long‐term trends in PWS have focused on alterations in canopy structure (e.g., leaf area index) or canopy structure‐dependent variables (e.g., gross primary productivity and evapotranspiration). This is largely because long‐term trends in canopy structure are relatively easy to detect from satellite observations. However, a focus on structural responses limits our ability to detect physiological stress due to challenges in isolating it from the effects of structural greening. Consequently, this difficulty hampers a comprehensive examination of long‐term PWS in the context of global greening trends. To address this gap, we developed a new process‐based metric for PWS to isolate physiological responses from structural greening, which we then used to detect global PWS trends over the past four decades. Combining site‐level and satellite observations at the half‐degree resolution across the globe, we found that accounting for greening‐related changes substantially alters the sign of long‐term PWS trends inferred from traditional approaches. Specifically, our study reveals a significant increase in PWS that is only detectable when accounting for structural greening trends. When greening trends are not accounted for, global PWS appears to have decreased over time. Overall, our results highlight the need to integrate structural dynamics and greening into PWS detection. Such an integration of observations and land models will improve our understanding of plant‐water‐energy interactions.
Field / plotLeafObject detectionStress / disease detectionDisease symptoms / severity
Plant diseases are generally caused by pest, insects, pathogens and decrease the productivity to large scale if not controlled within time. Agriculturists are facing lose due to various crop diseases. It becomes tedious to the cultivators to monitor the crops regularly when the cultivated area is huge that is in acres. The proposed system provides the solution for regularly monitoring the cultivated area and provides the automated disease detection using remote sensing images. The proposed system intimates the agriculturist about the crop diseases to take further actions. The objective of the proposed system is to early detection of diseases as soon as it starts spreading on the outer layer of the leaves. The proposed system works in two phases: the first phase deals with training data sets. This includes, training both healthy and as well as diseased data sets. The second phase deals with monitoring the crop and identifying the disease using Canny?s edge detection algorithm.
Cotton leaf diseases significantly affect crop productivity, fiber quality, and agricultural sustainability, making early and accurate disease diagnosis essential for effective crop management. Manual disease identification is time-consuming, subjective, and dependent on expert knowledge, highlighting the need for automated and intelligent diagnostic systems. This study proposes a CNN-based deep learning framework for automated multiclass classification of cotton leaf diseases using digital image analysis. The proposed framework is trained on a curated dataset comprising healthy and diseased cotton leaf images representing multiple disease categories. CNN architecture performs hierarchical feature extraction, automatically learning discriminative spatial representations from input images without manual feature engineering. The network is trained for 100 epochs using optimized hyperparameters to achieve robust feature learning while ensuring high generalization capability. Model performance is quantitatively evaluated using standard classification metrics, including accuracy, precision, recall, F1-score, and confusion matrix analysis, providing a comprehensive assessment of classification effectiveness. Experimental results demonstrate that the proposed framework achieves reliable disease recognition with high classification accuracy and minimal inter-class confusion, validating the effectiveness of CNN-based feature learning for agricultural image analysis. The automated system enables rapid and accurate disease diagnosis, supporting timely intervention, optimized crop protection strategies, and improved decision-making in precision agriculture. Future work will focus on improving model robustness through the integration of larger and more diverse datasets, advanced data augmentation techniques, transfer learning, and hybrid deep learning architectures. Furthermore, deployment of the proposed framework on mobile, edge, and web-based platforms, together with IoT technologies, can facilitate real-time field monitoring and disease surveillance. Overall, the proposed CNN framework provides a scalable, computationally efficient, and intelligent solution for automated cotton leaf disease classification, contributing to the advancement of AI-driven precision agriculture and sustainable crop management.
Abstract Chili is an important economic and nutritional crop with a relatively limited availability of different disease-resistant varieties. Leaf diseases, including those caused by fungi, bacteria, viruses, pests, or nutritional deficiencies, significantly compromise production and crop quality. Early detection of these diseases is key to reducing yield loss; however, traditional visual examinations are limited by time constraints, human subjectivity, and low detection sensitivity at early stages of infection. To overcome these challenges, this study proposes a deep learning–based framework for early multi-class detection of chili leaf diseases using convolutional neural networks (CNNs). A real-field dataset comprising 16,392 high-resolution images of chili leaves across six disease and healthy classes was collected from multiple regions of Bangladesh. The dataset was preprocessed, augmented, and split into training and validation sets using an 80:20 ratio. Five pre-trained CNN architectures—DenseNet121, EfficientNetB3, MobileNetV2, ResNet50, and InceptionV3 were evaluated using a transfer learning strategy. Experimental results demonstrate that MobileNetV2 achieved the best performance, attaining an overall classification accuracy of 96%. The results indicate that the proposed system demonstrates strong generalization capability and effectively discriminates visually similar chili leaf diseases. This work can be considered a valuable application in precision agriculture, providing an efficient, automated, and practical approach for in situ early diagnosis of chili leaf diseases through smart farm management.
Above ground crop traits provide an early indication of a plant's capacity to tolerate stress, and are important for breeding programs aimed at improving stress tolerance. In this work, we present a high-throughput methodology to study morphological and physiological traits of individual quinoa plants over time under control, drought, and saline conditions. We used daily sideview imaging of individual plants, followed by segmentation of the panicle, leaf and stem using the deep learning U-Net++ segmentation model. The resulting segmentations were used in regression models to estimate leaf area, fresh and dry biomass, and leaf dry weight. The regression models showed high predictive accuracy. Using these estimates, we could calculate specific leaf area and leaf weight ratio. In addition, radiation use efficiency for above-ground biomass production was calculated, providing an independent physiological check on the consistency of these predictions. Finally, using automated measurements of plant transpiration we were able to determine daily averages of whole plant stomatal conductance. The results show that image-derived morphological traits can be used to accurately estimate biomass-related traits and to derive physiologically meaningful indicators of plant performance over time. This method provides a framework for non-destructive monitoring of quinoa responses to drought and salinity.
Izabelle de Lima e Lima · Marta Laura de Souza Alexandre · Rodnei Rizzo · Ana Karla da Silva Oliveira · Carlos Augusto Alves Cardoso Silva · Peterson Ricardo Fiorio
Proper nitrogen (N) management is essential for increasing the productivity of sugarcane (Saccharum spp.) and reducing the economic and environmental impacts associated with excessive fertilizer use. This study compared the performance of two portable spectroradiometers, FieldSpec 3 and HandHeld 2, in estimating foliar nitrogen content based on hyperspectral data in the visible and near-infrared regions, obtained throughout the crop cycle. The experiment was conducted in Piracicaba, São Paulo, Brazil, under four N rates: 0, 60, 120, and 180 kg ha−1. Spectral measurements were taken at the foliar and canopy levels at eight evaluation times, accompanied by laboratory determination of N content. Partial Least Squares Regression (PLSR) and Random Forest (RF) models were fitted using the spectral data and days after cutting (DAC), included as a categorical factor and evaluated using 10-fold internal cross-validation, based on the metrics R2, RMSE, MAE, and Willmott’s refined agreement index (dr). The foliar data performed better with PLSR (R2 = 0.727; RMSE = 1.381 g kg−1; MAE = 1.109; dr = 0.917) than canopy data (R2 = 0.591; RMSE = 1.489 g kg−1; MAE = 1.157; dr = 0.866). PLSR also outperformed RF at both acquisition levels. The green (~550 nm) and red edge (~740 nm) regions were the most relevant for N estimation. Under the evaluated conditions, model performance was associated with the spectral acquisition level and conditions, the instrumental configuration, and the modeling strategy employed.
In India, to increase the wheat yield production and support sustainable agricultural practices, timely and correct identification of wheat crop diseases such as yellow rust, brown rust, etc., is very much essential. Further, the majority of deep learning algorithms show high efficiency in plant disease detection and recognition, but these studies rely on handcrafted datasets captured under controlled laboratory conditions and accordingly will not perform well in the real world domain. Furthermore, there is a lack of real field wheat crop disease datasets collected in India, particularly from Haryana state, being the largest producer of wheat crops. Therefore, to address the issue of lack of region specific crop diseases dataset, which are region-specific, the wheat crop disease data set is curated, consisting of 2672 images of healthy and diseased leaves of wheat crops collected from the real field of CCS Haryana Agricultural University, Hisar. The dataset includes typically three classes of leaves of wheat crop, i.e., Yellow Rust (910), Brown Rust (922) and Healthy leaves (840) collected using both a DSLR camera and different smartphone cameras under natural lighting and field conditions. All images were validated and labeled by taking the expertise of wheat pathologists to ensure reliability on the dataset. The study also provides a comprehensive, systematic workflow for transforming raw data into a high-quality benchmark dataset for training using image preprocessing and augmentation techniques. Besides this, a comparative benchmarking analysis is performed under identical experimental conditions using three widely adopted deep learning architectures, which includes ResNet50, MobileNetV2 and EfficientNet-B0. The experimental results show that ResNet50 and EfficientNet-B0 achieve similar performance i.e. approximately 91% classification accuracy on real-field data. However, MobileNetV2 offers a lightweight alternative suitable for mobile and edge deployment. Furthermore, Grad-CAM based explainability analysis was performed to validate model predictions and highlight disease specific regions in wheat leaves. Therefore, this study contributes a practical region specific WheatDisease-HRY dataset and baseline benchmarking framework for developing robust AI based wheat disease diagnosis tools for real world agricultural applications in India.
Billah MM, Sagor S, Hossain S, Alahmadi TJ, Moni MA, Uddin MS.
Field / plotLeafClassificationDisease symptoms / severity
Early and reliable diagnosis of hibiscus leaf diseases is critical to protect horticultural yield. Yet, it remains challenging under real-time field conditions where uncontrolled lighting, clutter, and the non-contiguous nature of pathological symptoms blur diagnostic cues. To address these challenges, we introduce CNN-FusionViT-GNN. This explainable hybrid multi-branch framework synergizes the fine-grained texture extraction of a DenseNet201 backbone, the global contextual modeling of a Vision Transformer (ViT), and the relational reasoning of a Graph Neural Network (GNN). The model is trained and validated on 'Hibiscus,' a curated field dataset of 1165 images from Bangladesh, which is strategically augmented to 8000 samples for robust training following a strict train-validation-test split. The proposed framework achieves a state-of-the-art accuracy of 98.33% with a macro F1-score of 0.98. The framework's generalization is confirmed through high performance on external datasets: 98.78% accuracy on the 52-class Plant City dataset and 83.88% on the 10-class Tomato Leaf Disease dataset, while maintaining a rapid inference time of 10-45 ms. Furthermore, a multi-faceted Explainable AI (XAI) audit using LIME, Grad-CAM++, ViT Attention Maps, and Occlusion Sensitivity validates that the model's decisions are driven by biologically meaningful symptom patterns rather than background artifacts. This study establishes a computationally efficient, transparent, and robust pathway for automated disease diagnosis in precision agriculture.
Reproduction assets foundThe paper's primary Hibiscus leaf disease image dataset is publicly deposited on Mendeley Data, and the external Tomato Leaf Disease dataset used for validation is also publicly available on Mendeley Data. No author analysis code or trained model checkpoints are reported.Dataset · publicThe primary dataset generated and analyzed during the current study,“Hibiscus Leaf Diseases Classification Dataset,”is publicly available in Mendeley Data 7 .Open asset ↗Mendeley Datalines:307-347Plant phenotyping relevance matchEurope PMC · checked 5 Sept 2026
Abstract Uneven occurrence of potato leaf diseases necessitates site-specific treatment rather than conventional uniform spraying. This study presents a vision-guided autonomous variable-rate spraying system integrating real-time deep learning-based disease detection, temporal disease severity estimation, and PWM-controlled precision spray actuation for targeted crop protection. A composite dataset comprising 1,561 field images containing 14,166 annotated leaf instances was developed under dense canopy conditions. Comparative evaluation of object detection architectures showed that YOLOv26 achieved the highest detection performance with a peak mAP@0.5 of 0.961, outperforming YOLOv8 (0.924) by 3.7% and YOLOv12 (0.938) by 2.3%. The lightweight YOLOv26n variant was selected for embedded deployment. When deployed on a Raspberry Pi 5, the optimized ONNX model achieved an inference speed of approximately 10 FPS, corresponding to a spatial sampling interval of 0.018 m at an operating speed of 0.18 m/s. A 2 s temporal sliding window generated a stable canopy-level disease severity index, which was mapped to PWM duty cycles for variable-rate pesticide application. Field experiments involving 598 spatial spray decision instances validated the proposed perception-driven spray control framework. Compared with conventional constant-rate spraying, the system reduced agrochemical consumption by 32.75% (from 458 to 308 L/ha). These results demonstrate the feasibility of integrating embedded deep learning and adaptive spray control into a field-deployable autonomous spraying platform for precision potato disease management.
Premise Accurate species identification is crucial for ecological restoration and can be especially challenging for understudied non-model species. Quercus garryana is the only native oak species in the Pacific Northwest and is an important component of the endangered oak savanna ecosystem. Quercus robur is an imported ornamental species from Europe and has been found to be mistakenly planted as Q. garryana in habitat restoration projects. Methods We measured leaf morphological traits sampled from herbarium collections in their native ranges using the digital morphometric tools MorphoLeaf and Tomato Analyzer. We then used Lasso logistic analysis to generate a predictive model and tested it on leaves from Portland, Oregon. To streamline this species detection process, we developed Garryanalyzer, an ImageJ plug-in that automatically measures leaf traits and outputs species predictions. Results Garryanalyzer demonstrated 95% accuracy in predicting the species identity of herbarium specimens of oaks. Garryanalyzer correctly identified all Q. robur individuals sampled in Portland but showed lower accuracy for Q. garryana . Discussion Many existing morphometric software are not open source, which makes them unable to be customized to specific study systems. Garryanalyzer is built upon the widely used open-source ImageJ platform. This study also demonstrates a viable workflow for developing similar tools for other ecologically important non-model plant species.
Reproduction assets foundThe paper's authors publicly released the Garryanalyzer ImageJ plug-in source code on GitHub, all original and modified leaf images used in the morphometric analyses on Zenodo, and the full leaf morphometric measurement dataset plus R Lasso analysis code in a second Zenodo repository. All are paper-specific, public,可直接Code · publicThe source code and installation instructions for Garryanalyzer can be accessed on GitHub at https://github.com/zxie8561/Garryanalyzer.Open asset ↗https://github.com/zxie8561/Garryanalyzer · zxie8561/Garryanalyzerhtml-lines:210-274Dataset · publicAll images used in the morphometric analyses, both original and modified, are available on Zenodo (https://doi.org/10.5281/zenodo.17462266).Open asset ↗https://doi.org/10.5281/zenodo.17462266 · 10.5281/zenodo.17462266html-lines:210-274Dataset · publicThe full dataset of leaf morphometric measurements of both GBIF and Portland samples, R code for Lasso analysis, and other miscellaneous files are available on a separate Zenodo repository (https://doi.org/10.5281/zenodo.17546152).Open asset ↗https://doi.org/10.5281/zenodo.17546152 · 10.5281/zenodo.17546152html-lines:210-274Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Komang Bagus Sadewa Netra · Putu Indah Ciptayani · Luh Gede Putri Suardani
Field / plotLeafClassificationStress / disease detectionDisease symptoms / severity
This study aims to develop a classification system for papaya leaf diseases based on digital image processing using the Convolutional Neural Network (CNN) method with the MobileNetV3 architecture. The background of this research is the manual process of identifying papaya leaf diseases by farmers, which is often inefficient and prone to misdiagnosis. The research method adopts a software engineering approach using the Agile development model, allowing iterative and flexible system development. The dataset consists of papaya leaf images categorized into three classes: curl, ringspot, and healthy, obtained from field observations and secondary datasets. The data were processed through preprocessing stages before being used to train the CNN model. The results indicate that the best model was achieved using a learning rate of 0.001 and 30 epochs, with a validation accuracy of 95.56% and a testing accuracy of 88.89%. The model demonstrates high confidence in classifying images, particularly for the curl and healthy classes. However, the confusion matrix reveals that the model's performance on the ringspot class remains relatively low due to a high misclassification rate. Overall, the developed system is capable of automatically identifying papaya leaf diseases and has strong potential for implementation as an Android-based application to support early detection and decision-making in plant care.
Potato leaf disease is one of the main problems in potato cultivation because it can reduce plant quality, decrease crop yield, and cause economic losses for farmers. Manual disease detection still has limitations because it depends on farmers’ experience and is prone to errors, especially when disease symptoms have similar visual characteristics. This study aims to apply the Convolutional Neural Network (CNN) algorithm to predict potato leaf diseases based on digital images. The dataset used in this study was obtained from Kaggle and consisted of 1,500 potato leaf images divided into three classes: healthy leaves, early blight, and late blight. The research stages included dataset collection, data splitting into training, testing, and validation data, CNN modeling using Jupyter Notebook, model training with 50 epochs, model evaluation using a Confusion Matrix, and model implementation into a web-based system using Flask. The test results show that the CNN model was able to classify potato leaf diseases with an accuracy of 97%. These results indicate that CNN is effective in recognizing visual patterns in potato leaf images, such as color changes, spots, and leaf damage. This study is expected to serve as a basis for developing an early detection system for potato leaf diseases that is faster, more accurate, and easier for farmers to use.
The early and effective diagnosis of tomato leaf diseases is very important to enhance crop yield and reduce economic loss in precision agriculture. The conventional image-based methods are typically based on single architecture model, which cannot capture fine-grained lesion details and global contextual patterns simultaneously in the real-field. To this end, we introduce a deep hybrid Convolutional Neural Network (CNN) –Transformer architecture by combining ConvNeXt Large (ConvNeXt-L) (as local feature extractor) and Swin Transformer (as global context encoder). The concatenated features vector is then fed to a shallow classifier to predict the disease. The model was tested on two datasets, namely a field dataset in agriculture areas from Madhya Pradesh (India) and a benchmark tomato leaf dataset. Experimental results revealed that the proposed scheme achieved accuracy of 92.83% on a primary dataset, and performance was significantly high with an accuracy of up to 95.65% in terms of generalization rate for computing technique models from various environmental conditions.
Reproduction assets foundThe paper uses a public Tomato Leaves Dataset from GTS AI as its secondary/external validation dataset for tomato leaf disease classification. The primary field dataset from Madhya Pradesh is not stated as publicly available, and no author analysis code or trained model is reported as deposited.Dataset · publicSecondary Dataset: The Secondary dataset was extracted from the public Tomato Leaves Dataset
available at GTS AI platform. It involves various disease classes, such as bacterial spot, early blight,
late blight, leaf mold, powdery mildew, septoria leaf spot and spider mites (Figure 1) target spots are
present in tomato mosaic virus leaves yellow curl virus of tomato. This data set was employed as an
external validation to evaluate the generalization of proposed model in different conditions and
diseases types.
Source : https://gts.ai/dataset-download/tomato-leaves-dataset/Open asset ↗GTS AIpdf-page:20 lines:1-23Code / dataset availability confirmedCrossref · checked 14 Sept 2026
Published23 Jul 2026Journal of Intelligent Decision Making and Information ScienceCited by 0 · OpenAlex ↗
- Enhancing agricultural productivity and attaining sustainable crop management depend on the early and precise identification of leaf disease. Using state-of-the-art technologies in precision agriculture like machine learning (ML) and image processing greatly increases the effectiveness of disease detection and facilitates well-informed decision-making. But conventional manual inspection techniques are still tedious, unpredictable, and prone to errors. In order to overcome these constraints, this research offers YOLOv12-CropNet, an innovative deep learning-based system for multi-crop leaf disease diagnosis in real time. The proposed YOLOv12-CropNet approach makes use of the Convolutional Block Attention Module (CBAM) for adaptive attention, the Content-Aware Reassembly of Features (CARAFE) up-sampling module to preserve fine-grained disease characteristics, the YOLOv12 architecture improved with Ghost Convolution for effective feature extraction, and Involution layers to capture spatially specific patterns. Inspection techniques are still laborious, arbitrary, and prone to mistakes. A substantial set of data of 38 classes of both healthy and sick leaves from a variety of crops, including tomato, potato, apple, grape, corn, mango and sugarcane, was put together for training and evaluation. Experimental results show that YOLOv12-CropNet finds a suitable balance between computational speed and accurate detection. Accuracy, F1-score, recall, and precision are important performance metrics that verify the model's resilience in challenging environmental and visual circumstances. The suggested technique provides a scalable and field-deployable way to assist effective identification of diseases and precision agricultural decision-making. The proposed YOLOv12-CropNet model exhibits better performance than the other evaluated models, attaining a 98.45% peak accuracy, 98.10% precision ,98.20 % sensitivity and a 98.18% F1 score, thereby highlighting its efficacy in multi-crop leaf disease detection.
Reproduction assets foundThe paper uses public Kaggle datasets as its phenotyping image inputs: the PlantVillage dataset (38 crop-disease classes) and the Sugarcane Leaf Disease dataset, both cited with explicit public URLs. No author code, models, or checkpoints are reported as publicly available.Dataset · public[37] PlantVillage Dataset. Available online: https://www.kaggle.com/datasets/abdallahalidev/plantvillage-datasetOpen asset ↗pdf-page:22 lines:1-61Dataset · public[39] Sugarcane leaf Disease Dataset available online: https://www.kaggle.com/datasets/nirmalsankalana/sugarcane-Open asset ↗pdf-page:22 lines:1-61Plant phenotyping relevance matchEurope PMC · Crossref · checked 5 Sept 2026
Abstract The diseases on sesame leaves have a huge implication on the production and earnings of farmers particularly in the developing areas. It is important to ensure that the disease is properly managed by identifying it early and correctly. The purpose of this study is to create a deep learning-based framework that could be used to categorize four of the most common scenarios involving sesame leaves, i.e. Healthy Leaf, Leaf Spot Disease, Yellowing Leaf Syndrome, and Leaf Damage by Insects using high-resolution images acquired in Pabna, Bangladesh. The preprocessing, augmentation and split of a set of 3,540 images were performed into training and validation sets. The pre-trained convolutional neural networks models were trained and tested on five models inceptionV3, EfficientNetB3, ResNet50, MobileNet, and DenseNet121 by measuring the metrics such as accuracy, precision, recall, and F1-score. MobileNet achieved the highest accuracy of 96.33%, precision of 96%, recall of 96%, and the F1-score of 96%, which is the best amongst them. The findings indicate that deep learning architectures are capable of classifying the sesame leaf diseases in a reliable and precision-oriented way that is not affected by different environmental circumstances. The study facilitates the creation of the automated, efficient methods of detecting the disease at an early stage, cutting down the number of pesticides used and enhancing crop control. Future direction will be to enlarge the dataset, add temporal data and to implement lightweight models so that it can be deployed to real-time field projects
Seedling establishment represents a critical phase in early crop growth and development, directly influencing biomass accumulation and yield potential. To characterise early growth dynamics under field conditions, both growth rate and uniformity of emergence need to be assessed continuously; however, manual quantification of these dynamic traits in large-scale trials remains impractical. Here, we present LeafTip-RN, an open-source and deep learning (DL)-powered pipeline for dynamically measuring wheat (Triticum aestivum L.) early establishment in the field. To enable flexible and scalable data collection, ultralow-altitude drone phenotyping was employed, followed by the development of an optimised DL model to automate leaf-tip-related feature extraction from complex backgrounds. Notably, to address data sparsity arising from eight phenotyping timepoints, we integrated an image-to-video generative AI (GenAI) module into the pipeline to interpolate keyframes between early and late seedling stages (i.e. 18-40 days after sowing), resulting in a training library comprising 353,019 labelled leaf tips. Using the pipeline, we successfully quantified multiple agronomically important establishment-related traits (e.g. plot-level leaf tips and seedling spatial uniformity), followed by deriving their growth curves for 51 wheat varieties across two growing seasons (2024-2026). After validating these LeafTip-RN-derived traits, we further computed varietal relative growth rates and uniformity indices, based on which the 51 varieties were classified into high-, medium-, and low-performance groups, revealing discrepancies between LeafTip-RN-derived classification (18-40 DAS) and manual assessment at 40 DAS when dynamic early performance was considered. Finally, to facilitate broad adoption by the plant research community, we developed an openly accessible graphical user interface (GUI) for non-expert users to visualise and analyse rapid seedling developmental changes. Taken together, our study provides a scalable GenAI-powered solution for evaluating seedling establishment in wheat, offering valuable tools for breeders and researchers to identify varieties with enhanced early growth vigour and emergence dynamics that are extensible to other cereal crops.
Introduction Safeguarding the yield and quality of field crops against abiotic stresses is critical for large-scale agricultural production and precision agronomy. This study addresses the challenges of detecting concealed fruit stress and overcoming label scarcity in unmanned aerial vehicle (UAV) multispectral monitoring, using blossom-end rot (BER) in processing tomatoes as a case study. Methods We developed a leaf-fruit synergistic Roll and Color Disease Index (RCDI), integrating fruit incidence with visible canopy phenotypic features to characterize the combined canopy-fruit stress status associated with BER. A three-level screening strategy was used to identify the optimal spectral feature set. A Multi-model Collaborative Cyclic Self-Training (MCC-ST) framework was subsequently developed to address the limited availability of severity-labeled samples. Results The combination of GRVI, NDVI, and SAVI was identified as the optimal spectral feature set, achieving stable within-dataset binary classification performance of approximately 97% in repeated cross-validation. Under 30 random-seed repeated stratified three-fold cross-validations, MCC-ST + DT and MCC-ST + RF achieved RCDI-based BER severity-grading accuracies of 85.00% ± 0.31% and 85.07% ± 0.42%, respectively. Compared with the corresponding original DT and RF models, MCC-ST improved repeated-validation accuracy by 16.19-4.09 percentage points. Discussion The RCDI helps bridge the observational gap between canopy signals and concealed fruit stress, while MCC-ST alleviates the bottleneck associated with label scarcity. The proposed approach provides a promising framework for crop abiotic-stress monitoring under limited-label conditions.
Chili plants (Capsicum sp.) are one of the important horticultural commodities in Indonesia with high economic value. However, chili productivity is often reduced due to leaf diseases such as leaf curl, yellow leaf virus, and leaf spot disease. Manual disease identification conducted by farmers still has several limitations because it requires considerable time, experience, and is prone to observation errors. Therefore, an automatic system is needed to support early detection of chili leaf diseases quickly and accurately. This study aims to develop a chili leaf disease classification system using a Convolutional Neural Network (CNN) based on leaf images. The dataset used in this study consists of chili leaf images categorized into four classes, namely healthy leaves, leaf curl, yellow leaf, and leaf spot. The research stages include dataset collection, image preprocessing, data augmentation, CNN model training, and model evaluation using a confusion matrix with performance metrics including accuracy, precision, recall, and F1-score. The results show that the CNN model is capable of classifying chili leaf diseases with satisfactory performance. Based on the evaluation results, the model achieved a precision value of 0.75, a recall value of 0.53, and a mean Average Precision (mAP@0.5) value of 0.60. The developed system is also able to display classification results along with the confidence score of the prediction. Therefore, the CNN method has strong potential to be implemented as an image-based early detection system for chili leaf diseases to assist farmers in monitoring plant conditions more effectively.
Abstract Outbreaks of plant diseases are major threats to world food security particularly in areas where real-time monitoring and quick decision support are constrained by low-power edge gadgets and untrustworthy connectivity. In order to overcome these issues, this paper presents a FPGA-Accelerated IoT implementation of a Causal-Attention Multi-Modal Deep Learning Network, named EpiFusionNet-Edge, that can be applied to monitor crop diseases with real-world farming scenarios with high precision and scalability. The framework incorporates five data modalities that are complementary in nature and they include RGB leaf pictures, microscopic foldscope images, UAV hyperspectral signatures, microclimate IoT sensor measurements and region-specific pathogen/pest pressure indexes giving a complete picture of the health of the plant. Dual causal-attention mechanism is proposed to simulate both spatial and temporal environmental factor activation, which helps to detect and make predictions at the early stage and provides an explanatory logic behind the decisions. Multi-task learning enables classification of diseases, quantification of their intensity at the level of a micro-prediction and prediction of outbreaks in the short term (1–30 days). In order to achieve deployability in resource-constrained settings, the proposed deep learning architecture is ensemble-distilled, structurally pruned, and INT8-quantized, and hardened on a Xilinx Zynq-7000 FPGA platform. The FPGA accelerator is 43.2x faster inference, 88 percent less power usage, and less than 10 ms latency, which allows real-time execution of continuous field monitoring with IoT sensors. Cross-condition assessment on multi-domain datasets shows that there are great improvements on cross-environment generalization rates with 98.6% classification accuracy, 92.7% severity estimation accuracy and less than 3.5% degradation with domain shift. Grad-CAM + + and causal feature traceability further add interpretability with the focus of the model and the pathological indicators proven by experts. The findings show the promise of using a combination of IoT sensing, multi-modal AI fusion, and FPGA hardware acceleration to develop a deployable and scalable and transparent system with regard to precision agriculture. This paper creates a roadmap to a new generation of smart farming systems that are able to conduct disease surveillance and actively protect crops at the periphery in an autonomous manner.
ABSTRACT Banana diseases impose severe production losses in tropical smallholder farming systems, yet accurate in-field visual diagnosis remains difficult: symptom expression varies across cultivars and growth stages, and several diseases produce morphologically overlapping foliar signs. We developed a probabilistic image-recognition framework for detecting five economically important banana diseases — Xanthomonas Wilt, Banana Bunchy Top Disease, Fusarium Wilt (Panama disease), Yellow Sigatoka, and Black Sigatoka — from in-field photographs, without any disease-specific fine-tuning of the vision backbone. The approach extracts frozen 1,152-dimensional embeddings from the DINOv3 vision foundation model and couples them with a conditional normalizing flow, trained on four publicly available datasets spanning diseased banana plants, healthy tissue, non-banana vegetation, and general natural imagery. On an independent test set the model achieved F1 scores exceeding 0.98, average precision values of 0.968–0.999, and AUROC values of 0.997–1.000 across all five diseases evaluated as binary detection problems. Multi-class accuracy was near-perfect, with limited confusion between Yellow Sigatoka and Black Sigatoka — a biologically plausible ambiguity attributable to overlapping early-infection foliar symptoms. Because the normalizing flow estimates explicit conditional probability densities rather than decision boundaries, two complementary log-likelihood ratios can be derived: a disease ratio comparing each disease class against healthy banana, and a plant ratio comparing banana against non-banana imagery. Together these define an interpretable two-dimensional diagnostic space that simultaneously quantifies evidence for disease presence and image relevance, cleanly separating diseased plants, healthy plants, and out-of-distribution images while flagging uncertain predictions for confirmatory testing. Inference on frozen embeddings is lightweight and compatible with smartphone deployment, providing a scalable, uncertainty-aware diagnostic tool for smallholder farming systems and disease surveillance programmes.
Abstract Blanket-rate agrochemical scheduling — a practice wherein the same quantity of fertilizer or pesticide is spread uniformly across an entire field irrespective of spatial or temporal crop need — persists as the dominant farm management paradigm across rural India and large parts of South Asia. This approach generates cascading inefficiencies: excess nitrogen drains into waterways, off-target pesticide deposits devastate pollinators, input costs erode thin profit margins, and wide-scale greenhouse gas release from soil microbial activity accelerates climate change. The study documented here addresses this challenge through a purpose-built, four-layer intelligent field management platform. The platform ingests continuous data from drone-mounted multispectral cameras, in-field IoT soil probes, a wireless weather station, and cloud-sourced Sentinel-2 satellite imagery, then passes these inputs through a cascaded AI inference stack. A fine-tuned YOLOv8-L network performs real-time pest and foliar disease localisation; a ResNet-50 backbone quantifies canopy health across five stress gradients; a two-layer stacked LSTM projects short-horizon yield trajectories; and a Deep Q-Network autonomously plans drone spray routes weighted by field-specific prescription maps. Field validation spanned two consecutive growing seasons (Rabi 2022–23 and Kharif 2023–24) across six georeferenced plots covering 4.8 ha at Baramati, Maharashtra. Outcome metrics recorded during head-to-head comparison with conventional practice included a disease detection score of 95.6% mAP, a 47.3% reduction in total nitrogen applied, a 38.1% decrease in pesticide volume, and a 22.4% uplift in harvested grain weight. Together, these field-verified numbers substantiate the operational readiness of integrated AI precision agriculture for smallholder deployment.
Reproduction assets found保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。Code · publicData Availability The annotated image dataset (14,300 images, 23 classes), trained YOLOv8-L and ResNet-50 weights, LSTM model files, DQN policy checkpoint, and all analysis scripts are archived at https://github.com/precision-agri-ai (Zenodo DOI: 10.5281/zenodo.XXXXXXX).Open asset ↗precision-agri-ailines:161-182Plant phenotyping relevance matchCrossref · checked 14 Sept 2026
Published21 Jul 2026International Journal of Drug Delivery TechnologyCited by 0 · OpenAlex ↗
Plant diseases are crucial for improving crop yield and ensuring sustainable agricultural practices, particularly for staple crops such as groundnut and paddy leaf. However, existing methods often suffer from limited feature discrimination, inadequate attention to disease-affected regions, and reduced performance under real-world conditions. To address these limitations, this research introduces a novel deep learning (DL)-based GOPI-NET framework for precise groundnut and paddy leaf disease classification. Initially, the input leaf images are enhanced using Bilateral Filtering (BF) and Contrast Limited Adaptive Histogram Equalization (CLAHE) to reduce noise and improve contrast. Subsequently, HSV color space segmentation is employed to precisely isolate disease-affected regions. The proposed Dual Attention Network (DuAtNet) integrates channel and spatial attention mechanisms within a ConvNeXt backbone to capture discriminative disease-specific features. An efficient Fuzzy Extreme Learning Machine (FELM) classifier is then utilized for final categorization into Healthy, Leaf Spot, Bacterial Wilt, and Leaf Blight classes. The effectiveness of the GOPI-NET is evaluated using precision, recall, specificity, accuracy, and F1-score. The experimental results demonstrate that GOPI-NET achieves an overall accuracy of 98.32%. The GOPI-NET improves classification accuracy by 1.29%, 1.40%, and 2.23% compared to GLDICCNN, DNN-CSA, and LeafNet respectively.
Jin Xu · Sina Jamalzadegan · Yiyang Shen · Yi Chen · Amanda C. Saville · Zach Hetzler · Aditi Dey Poonam · Zezhao Li · Mingzhuo Li · Ali Akbari · Mohammadreza Zare · Weixin Zhou · Oindrila Hossain · Xiangwu Zhang · Jean B. Ristaino · Alper Bozkurt · Yong Zhu · Qingshan Wei
TomatoMultimodalLeafClassificationDisease symptoms / severityStress response / tolerancePlant / canopy temperature
Wearable plant sensing systems for simultaneous biochemical and physiological monitoring with real-time multimodal data analysis remain limited. Here, we present FolioClip, a multimodal wearable patch that continuously monitors leaf temperature, humidity, light, CO 2 , and three volatile organic compounds (VOCs) with high selectivity. Its bookmark-inspired design enables secure attachment to plant leaves of diverse morphologies and it integrates a flexible printed circuit board for wireless data transmission. We also develop FolioOmni, an open-source machine learning (ML) framework for sensor importance ranking, multi-stress classification, and early stress detection. The integrated FolioClip–FolioOmni platform detects and classifies nine stresses, including light, water, CO 2 , mechanical cut, P. infestans , and A. alternata , in tomato plants with 92% accuracy. Notably, P. infestans on tomato was detected within 15.5 h post-inoculation, earlier than quantitative polymerase chain reaction (qPCR) (~4 days) and visual phenotyping (~7 days), highlighting the potential of integrating multimodal wearable sensing and online ML for precision agriculture.
José M. R. Flauzino · Abdulkadir Sanli · Rudolph R. Shirima · Yuanjun Cai · Tinghao Hu · Leyang Li · Taliya Weinstein · Hong Seok Lee · Tayeb Tadlaoui · Selin Olenik · Abdülkadir Gümüşçü · Laura Gonzalez-Macia · George Mahuku · James Legg · Anthony E. G. Cass · Firat Güder
Cassava brown streak disease (CBSD) threatens food security for millions in East Africa, yet its control remains limited by the absence of field-deployable molecular diagnostics. Here, we introduce ELLA (Electrochemical Lateral flow assay with Linked Analytics), a battery-free, smartphone-powered electrochemical lateral flow assay that delivers enzyme-linked immunosorbent assay (ELISA)-grade protein detection directly in the field. ELLA integrates near-field communication, a single-chip potentiostat, metal-pin electrodes, and ferrocene-labeled nanoparticles into a fully disposable cassette, enabling quantitative immunoassays without optical instrumentation or centralized laboratory infrastructure. Validated across laboratory studies and extensive field trials in Tanzania, ELLA achieved 95% agreement with ELISA and 89% agreement with RT-qPCR, outperforming ELISA’s limit of detection while maintaining a material cost below US$1. By coupling molecular test results with cloud-linked analytics, we further trained DeepELLA, a smartphone-based image classification model that enables scalable surveillance from field-acquired leaf images. Together, these advances unify electrochemical sensing, digital connectivity, and AI-assisted interpretation, enabling portable, ELISA-level diagnostics for plant, environmental, and health monitoring in resource-limited regions.
Plant diseases significantly affect agricultural productivity and global food security, while accurate disease identification remains challenging because of uncertain and overlapping visual symptoms in leaf images. Existing deep learning approaches often require large annotated datasets and suffer from limited interpretability in practical agricultural environments. This study presents a multi-parameter improved fuzzy soft set-based framework for plant disease classification using tomato leaf images from the PlantVillage dataset. The objective is to develop an interpretable and reliable classification model capable of handling uncertainty in plant disease patterns through feature-driven fuzzy similarity analysis. The methodology integrates image preprocessing, color and texture feature extraction, variance-based feature weighting, prototype generation using K-means clustering, and fuzzy similarity computation using Mahalanobis distance and Gaussian membership functions. RGB, HSV, and Gray-Level Co-occurrence Matrix (GLCM) features are extracted from standardized leaf images and evaluated within an improved fuzzy soft classification framework. Performance comparison is carried out using machine learning models including Support Vector Machine (SVM), Random Forest (RF), Linear Discriminant Analysis (LDA), and Naive Bayes (NB) implemented in Python using Scikit-learn libraries. Experimental simulation results demonstrate that the proposed framework achieves competitive classification performance while preserving interpretability and robustness under uncertain feature distributions. Performance evaluation is conducted through accuracy analysis, ROC-AUC curves, confusion matrices, ablation studies, and Wilcoxon Signed-Rank statistical testing. The proposed Improved Fuzzy Soft model achieved an accuracy of 88.57% which is less than LDA (94.92%), Random Forest (97.78%) and SVM (97.94%) classifiers. However, in the cross data set validation, the proposed Improved Fuzzy Soft model achieved an accuracy of 67.35% which is greater than LDA (51.02%), Random Forest (51.02%) and SVM (55.10%) classifiers. Statistical validation using the Wilcoxon Signed-Rank Test produced a p-value of [Formula: see text], confirming that the performance difference between the Improved Fuzzy Soft framework and the Random Forest classifier is statistically significant under the current experimental setting.
Reproduction assets foundThe paper uses public tomato leaf image datasets (PlantVillage and PlantDoc from Kaggle) as phenotyping inputs and states the authors' Improved Fuzzy Soft Framework implementation is publicly available on Zenodo with source code and reproduction instructions.Dataset · publicThe dataset analyzed during the current study are available in the repository:
https://www.kaggle.com/datasets/abdallahalidev/plantvillage-datasetOpen asset ↗kaggle.com/datasets/abdallahalidev/plantvillage-datasetpdf-page:24 lines:1-75Code · publicThe implementation of the proposed Improved Fuzzy Soft Framework is publicly available through the Zenodo repository:
https://doi.org/10.5281/zenodo.20570546
The repository contains the source code, documentation, and instructions required to reproduce the experiments reported in
this study.Open asset ↗zenodo · 10.5281/zenodo.20570546pdf-page:25 lines:1-74Plant phenotyping relevance matchCrossref · Europe PMC · checked 16 Sept 2026
Abstract Detecting plant leaf diseases at an early stage is one of the most important requirements for sustainable agriculture, increasing crop productivity, and achieving the global Sustainable Development Goals (SDGs). However, accurately recognizing them in real-world farm fields can still be difficult due to factors such as background complexity, changes in light conditions, and very similar looking classes from a visual standpoint. In order to solve these problems, the authors here present a new Multi-Scale Feature Fusion (MSFF) model that can offer robust and highly accurate performance in identifying plant leaf diseases. Firstly, the brand new hybrid method starts with a U-Net segmentation designed exclusively to separate the diseased parts and thus allow the classification to be more robust. Next, Rank Order Fuzzy (ROF) is implemented to get rid of the background while still maintaining the edges, and the additional data is used for the network to generalize better. The color distribution is then analyzed to determine the variations in color brought about by the infection. In terms of features, EfficientNet is paired with an Attention-based Autoencoder to produce both spatially detailed global features and compact latent representations. The two sets of features are then combined through Canonical Correlation Analysis (CCA) which not only identifies the dependencies between the features but also enhances the discriminative strength. The final fused feature set is fed to module based on YOLO for detection and classification in order to obtain the final result of the plant disease identification system which is both accurate and fast. Among other datasets, the model has been tested on different apple leaf datasets such as FGVC7, AppleLeafSet, PlantVillage Apple, Kaggle Apple Leaves, and ATLDSD, which contain five disease classes. The results of the experiments indicate a classification accuracy of 99.95%, thus the model is superior to several state-of-the-art deep learning and classical machine learning methods. Statistical methods like five-fold cross-validation, paired t-tests ( p
Plant leaf diseases cause severe losses in crop yields and qualities, and account for considerable volume of losses to the agricultural output globally. Recognition of plant disease early and rightly is crucial to disease treatment and to reduce loss to the crop and to maintain agricultural sustainability. Plant disease that occurs on the leaves has been traditionally detected by farmers and experts with naked eyes by checking its symptoms like discoloration, spots and lesions. However, the process requires time, labour, expertise and is subjective, which renders it unusable for large-scale implemented agriculture. Recent years have seen the promising use of Artificial Intelligence (AI) as a tool for automated plant disease identification. The extraction of manually-crafted features from photographs of plant leaves, such as colour, texture, and form, is at the heart of many Machine Learning (ML) approaches used for disease classification. While these ML models have shown acceptable performance, they require significant manual feature engineering and can be poor at operating in real-world settings and with voluminous data. To address these issues, Deep Learning (DL) algorithms have found extensive usage in the identification and categorisation of plant leaf diseases. The You Only Look Once (YOLO) family of detection of objects models is making waves in the DL object detection space thanks to its impressive dual-tasking capabilities: object identification and multiple illness categorisation in a single pass, all at lightning speed and with pinpoint accuracy. For real-time disease identification in precision agriculture, YOLO stands out as an end-to-end feature learning and object recognition method, set apart from typical ML approaches. Understanding the DL models suggested for plant leaf disease detection and classification using the YOLO principle is the primary goal of this survey. It also provides a comparative and performance analysis of these models by examining their techniques, merits, demerits, datasets used, and evaluation metrics.
Accurate crop stress detection is essential for precision agriculture; however, most existing approaches rely on binary labels that collapse distinct stress processes water deficit, nutrient deficiency, disease, and pest damage into a single "stressed" category.We demonstrate empirically that this binary formulation is the primary barrier to classification performance: five model architectures achieve ROC-AUC values within ±0.01 of the random baseline (0.50) on binary stress classification, regardless of feature engineering strategy. Decomposing the binary label into stress-type-specific categories enables anXGBoost classifier to achieve 91.4% accuracy and a macro-averaged F1-score of 0.93 using the same underlying features.To extend coverage to visual disease symptoms, we train a MobileNetV2-based CNN on paddy leaf images, achieving 93.7% binary accuracy (healthy vs. disease_stress) with 100% healthy recall.We combine both modalities in a fusion ensemble that merges tabular and image predictions through rule-based priority logic, achieving 94.6% accuracy on the evaluated image subset.
Grapevine yellows diseases, including Flavescence Dorée (FD) and Bois Noir (BN), severely affect vineyards, making reliable detection methods essential. Detecting symptoms, particularly in Chardonnay variety, remains challenging due to year-to-year vine variability and the presence of symptoms that can be misleading. We present a dataset of multispectral images of grapevine leaves collected over four years (2021–2024). The dataset includes healthy leaves, grapevine yellows - overwhelmingly BN, and three other diseases (Esca, Discoloration, and Leafroll) with visually similar symptoms particularly for Leafroll, providing a diverse benchmark for disease recognition. Ground-truth labels were assigned based on field inspections conducted by Comité Champagne expert viticulturists who selected the vines before acquisitions and analyzed the images after acquisition. Images were acquired under laboratory conditions using a DJI Phantom 4 Multispectral camera from leaves collected in the Plumecoq Comité Champagne experimental vineyards. Leaves were collected on designated vines and placed on polystyrene boards on a 2x2 grid. One RGB image and five multispectral images corresponding to the Blue (B - 450 nm ± 16 nm), Green (G - 560 nm ± 16 nm), Red (R - 650 nm ± 16 nm), Red-Edge (RE - 730 nm ± 16 nm), and Near-Infrared (NIR - 840 nm ± 26 nm) bands were captured using the same acquisition system (camera-to-board distance of approximately 90 cm) over the four years.. For the vast majority, ambient light was used; for some acquisitions, indirect halogen spotlights were used. The 512x512 leaf images were manually cropped from the acquired original images with the same offsets between the bands; offsets were already present due to the parallax phenomenon. The dataset is organized into six folders (B, G, R, RE, NIR, and RGB). Each leaf is represented by one RGB image and five corresponding multispectral band images, enabling multimodal analysis and machine learning for grapevine disease detection. Each image filename encodes acquisition and annotation information as follows: Characters 1–2: acquisition year (21 = 2021, 22 = 2022, 23 = 2023, 24 = 2024). Character 3: image type/band (0 = RGB, 1 = B, 2 = G, 3 = R, 4 = RE, 5 = NIR). Character 4: class (J = Grapevine Yellows, T = Healthy, S = Esca, E = Leafroll, D = Discoloration). Character 5: illumination condition (0 = ambient light, G = additional left halogen spotlight, D = additional right halogen spotlight, 2 = additional both left and right halogen spotlightsl). Characters 6–9: image identifier, numbered sequentially from 0000. Character 10: relabelling status. After image acquisition, all leaves were independently reviewed by expert viticulturists on the recorded RGB images. A value of 1 indicates that the expert confirmed the original field label, 0 indicates that the original label was considered incorrect based on the information visible in the RGB image, and D indicates that the sample was reclassified as Discoloration. This dataset complements the "Multi-annual spectral data of Chardonnay grapevine leaves" dataset published on Recherche Data Gouv. While the previous dataset contains spectral measurements, the present dataset provides multispectral images acquired from the same grapevine leaves, enabling multimodal analyses that combine spectral signatures with image-based information. Note, however, that there is no bijection between the respective files.
Abstract Understanding the link between genetic variation and observable traits is key to crop breeding. Hyperspectral imaging captures physiological and biochemical profiles, but current supervised methods require costly trait annotations and treat each observation as a static snapshot, ignoring the temporal dynamics of plant development. We introduce SST-MAE, a self-supervised framework that learns genotype-discriminative representations from plant hyperspectral developmental trajectories, without requiring phenotypic labels. The model learns to reconstruct masked information, capturing multiple growth trajectories. Validated on 194 field-grown lettuce genotypes across eight time points, the frozen encoder serves as a feature extractor for downstream genotype classification. SST-MAE outperforms raw spectral and linear baselines, achieving AUROC > 0.89 for anthocyanin pigmentation SNPs and 0.77 for leaf serration. The learned features are highly label-efficient, attaining near-full performance with only 30–50% of labeled data, offering a scalable pathway toward high-throughput genetic screening from image-based phenotypes.
Plant disease detection under complex climatic conditions and cross-crop scenarios remains a critical challenge. To address this, we propose a novel SimAM-Inception-StyleRandomization (SIS)-YOLOv11 algorithm based on YOLOv11n for early/late blight detection on potato and tomato leaves. Our core innovations are: 1) A C3k2-SSI module integrating Style Randomization, Inception architecture, and SimAM attention to enhance cross-crop generalization; 2) A Fusion-InceptionConv module for fine-grained feature extraction under rainfall/haze noise; 3) SPPF-Inception and C2PSA-IS modules to optimize multi-scale feature fusion; 4) DepGraph pruning to reduce 47.82% parameters while improving performance. Experiments show that the pruned SIS-YOLOv11 outperforms YOLOv11n by 3.7% in precision, 6.6% in recall, 5.4% in mAP50, and 7.9% in mAP50-95, and surpasses mainstream models (Faster R-CNN, SSD, etc.). This study provides a robust, lightweight solution for automated cross-crop disease detection in complex agricultural environments.
Reproduction assets foundThe paper's image dataset (potato/tomato leaf disease images with annotations and climate-noise augmentation) is explicitly declared publicly available on Baidu AI Studio. No author code or trained model deposit is stated.Dataset · publicData Availability: All image datasets used and analyzed in this study are publicly available from the Baidu AI Studio dataset repository at the URL: https://aistudio.baidu.com/datasetdetail/245434 .Open asset ↗Baidu AI Studio · 245434lines:1-133Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Context: Early identification of plant diseases plays a crucial role in enhancing crop productivity and promoting sustainable agricultural practices. Advances in artificial intelligence and web-based technologies have paved the way for smart systems capable of automatically diagnosing diseases in crops like rice and sugarcane. Objective: This research focuses on developing a smart plant disease diagnosis system that integrates deep learning techniques with a MERN (MongoDB, Express.js, React.js, Node.js) stack to provide accurate, real-time classification of rice and sugarcane leaves diseases through a user-friendly web interface. Method: The proposed framework employs a Convolutional Neural Network (CNN) built with PyTorch and trained using a carefully curated dataset of diseased rice and sugarcane leaf images. The developed model was incorporated into a web application built using the MERN stack to enable seamless frontend-backend communication and real-time disease prediction. The model’s effectiveness was assessed using evaluation metrics such as precision, recall, F1-score, and confusion matrix analysis. Results: The CNN model achieved high classification performance, with an average class accuracy of 95.92%, overall classification accuracy of 91.83%, average precision of 91.85%, average recall of 92.05%, and average F1-score of 91.86%. Confusion matrix analysis further validated the model’s efficiency in accurately recognizing rice and sugarcane leaves diseases. The integrated web platform demonstrated efficient and user-friendly real-time disease prediction capabilities. Conclusions: The developed AI-based plant disease detection system highlights the effectiveness of integrating deep learning techniques with modern web technologies to support scalable agricultural solutions. The system provides a practical solution for farmers and agronomists seeking early and accurate crop disease detection. Future enhancements may include multilingual support, mobile application integration, and agronomic advisory modules to further advance precision agriculture initiatives.
The assessment of water stress levels in plants should be essential part of precise irrigation management, and a good and quick method is useful in plant phenotyping. There are many options for this task, but the effectiveness varies between crop species and environments. Apparently open field applications face the most difficulties. This study aimed to test a large number of vegetation indices (VIs) and multivariate models based on hyperspectral reflectance data (325-1075 nm) regarding their correlation and prediction abilities to leaf stomatal conductance, relative water content (RWC), and detailed chlorophyll, and carotenoid components. Data of the abovementioned variables was collected during three consecutive growing seasons in processing tomato cultivated under different water supply regimes to provide data with varying water stress levels. Then the relation of the measured variables to 226 VIs was tested created according to the formulas collected in the Index DataBase (IDB Project, indexdatabas.de ). New VIs were also developed derived from the most important variables of the ML algorithms, customised to tomato water stress assessment. Standard normal variate and its combination with Savitzky-Golay first derivative were used for pre-processing the spectra and principal component regression (PCR), partial least squares regression (PLSR), elastic net (ENET), support vector regression (SVR), random forest (RF) and extreme gradient boosting (XGB) algorithms were tested. The newly developed indices outperformed the existing formulas, except in the case of β-carotene. The most reliable index was developed for RWC estimation; that was the difference of the reflectance on the 986 and 701 nm wavelengths. The ENET and SVR algorithms produced the best models depending on the pre-processing method. The blue, near-infrared (NIR) and green regions, respectively, were the most important regarding all models according to the variable importance analysis. The model with the best metrics was developed for chlorophyll-a (R 2 =0.82, nRMSE=11%, RPIQ=2.41), followed by RWC (R 2 =0.72, nRMSE=14%, RPIQ=2.53).
Accurate and reliable diagnosis of grape leaf diseases is essential for sustainable viticulture, enabling timely intervention, reducing economic losses, and supporting precision crop management. Although deep learning (DL) models have demonstrated remarkable classification performance, their reliability under real-world field conditions remains insufficiently explored. In particular, confidence estimates often fail to reflect true predictive correctness when models are exposed to distributional shifts, limiting their practical applicability. To address this challenge, this study proposes a confidence- and uncertainty-aware DL framework for grape leaf disease diagnosis that extends evaluation beyond conventional accuracy-based metrics. Two publicly available grape leaf datasets were employed for stratified five-fold cross-validation in binary and multiclass classification tasks, while a third independently collected dataset was reserved exclusively for leakage-free external validation. EfficientNet-B0 and MobileNetV3-Large were evaluated under four inference configurations: raw inference, temperature scaling (TempScaling), Monte Carlo dropout, and an ensemble strategy combining uncertainty estimation with calibration. External validation demonstrated that both architectures maintained discriminative capability under domain shift. Under ensemble inference, EfficientNet-B0 achieved an accuracy of 73.8%, a macro-F1 score of 73.3%, a Matthews correlation coefficient (MCC) of 0.468, and a receiver operating characteristic area under the curve (ROC-AUC) of 0.804, while MobileNetV3-Large achieved 71.8% accuracy, 71.2% macro-F1, an MCC of 0.429, and a ROC-AUC of 0.787. Despite these promising results, raw predictions exhibited substantial overconfidence, with Expected Calibration Error (ECE) values of 0.287 and 0.312 for EfficientNet-B0 and MobileNetV3-Large, respectively. TempScaling markedly improved calibration quality, reducing ECE to 0.038 and 0.042 without affecting classification performance. Ensemble inference further enhanced the balance between predictive discrimination and reliability. The results demonstrate that strong classification performance alone is insufficient for trustworthy deployment in agricultural environments. Confidence calibration and uncertainty quantification provide complementary information for identifying overconfident predictions and improving decision reliability under field variability. The proposed framework offers a reliability-oriented approach for developing trustworthy artificial intelligence systems for grape leaf disease diagnosis in precision agriculture.
Rising temperatures and changing weather conditions are accelerating the spread of plant diseases and increasing the threat to global food security. Reliable detection of leaf diseases is therefore essential to protect crop yields and ensure food quality. Deep learning has proven to be a powerful tool for classifying leaf diseases across various crops. Due to the natural variability of plants, plant diseases often appear in irregular structures. Surface unevenness, folds, or dirt particles are common in field images and can be mistakenly identified as important features by convolutional neural networks (CNNs). This is a challenge that has not been sufficiently addressed in previous studies. This study proposes a novel deep learning approach that takes into account both the specific visual characteristics of plant diseases and potential disturbances in the microstructure, such as surface irregularities or prominent leaf veins, which may mislead the model. Using stratified five-fold cross-validation on a peer-reviewed dataset, which comprises 2,801 images of radish leaves across five classes (healthy, three disease classes: mosaic virus, black leaf spot, and downy mildew, and one pest-affected class: flea beetle), the proposed method achieved an average and balanced accuracy of 99.86%, establishing a new dataset-level benchmark in the field and demonstrating its effectiveness. The results indicate that the proposed approach may provide a promising basis for future applications in agricultural field monitoring, automated sorting and post-harvest quality control, offering potential to reduce both food waste and associated costs.
Whole plant leaf expansion (shoot expansion) under drought drives the trade-off between water saving for later grain production and canopy photosynthesis. Fine-tuning shoot expansion could therefore become a target of genetic progress for drought-prone environments. However, its components (axis production, i.e. tillering, leaf production on each axis, and individual leaf elongation) may have their own genotypic variability and plasticity under drought, making hard to calibrate crop simulation models and specify breeding targets. In this study, we focused on the genetic diversity of bread wheat and durum wheat to determine the links and trade-offs between the underlying processes of shoot expansion under drought and how it translates at the whole plant and canopy level. For that, we used non-destructive imaging both in the field and controlled condition platforms to determine their dynamics and analyze their relative contribution to the genotypic variability of whole-plant shoot expansion under drought. Results show that shoot expansion measured at plant level in controlled environment was associated with that measured at canopy level in the field, indicating that controlled phenotyping platforms can capture the genotypic variability of growth in the field. Both whole-plant and canopy expansion were associated with tillering rate. In addition, the sensitivity of shoot growth and tillering to soil water deficit were correlated, indicating that both tillering ability and sensitivity to water deficit drive the genotypic variability of shoot expansion. Overall, dissecting shoot- expansion dynamics allowed determining the links between shoot expansion traits under drought, and provides key targets in phenotyping, modelling and breeding for drought environments.
Early detection of heavy metal stress in plants is essential for effective environmental monitoring, particularly in contaminated urban areas. This study evaluated whether remote sensing combined with simplified anatomical diagnostics can provide a rapid and reliable method for detecting chromium (Cr) and nickel (Ni) stress in common urban weed species. Five species were selected: Trifolium pratense, Rumex acetosa, Alcea rosea, Amaranthus retroflexus, and Plantago lanceolata. Visible plant injuries were assessed using Evans Blue staining and image-based anatomical analysis, which enabled distinguishing between living, partially damaged, and dead cells. Multispectral observations using a MicaSense RedEdge-M camera allowed calculation of the Normalized Difference Vegetation Index (NDVI) to detect stress-related changes in photosynthetic apparatus. The studied species differed in their capacity to accumulate and translocate Cr and Ni. Metal bioaccumulation was low in all species (bioconcentration factor < 1), with the highest Ni accumulation observed in Plantago lanceolata. Translocation of both metals was the greatest in Trifolium pratense and Amaranthus retroflexus. Hydrogen peroxide levels increased in roots and leaves of all species, particularly in Alcea rosea. Despite the absence of visible injuries, microscopic anatomical changes were detected in T. pratense and R. acetosa, while NDVI values differed between sites. In summary, this study indicates that no simple relationship was found between physiological stress parameter values and NDVI. It is important to emphasize the need for continued research under controlled conditions with specific doses of PTEs salts. This should clearly demonstrate the relationship between plant physiological responses to stress and the results of multispectral observations.
Fusarium head blight (FHB) is a major mycotoxigenic disease of wheat, causing yield and quality losses and deoxynivalenol contamination. Rapid, non-destructive tools are needed to detect FHB, monitor wheat physiological responses, and evaluate sustainable management strategies, including biological control agents. Although vegetation spectroscopy is widely used for high-throughput phenotyping, most spectral studies focus on binary disease detection, while the capacity of hyperspectral data to capture concurrent host–pathogen–biocontrol responses across leaf and canopy scales remains underexplored. Here, we tested a full-range (400–2400 nm) hyperspectral phenotyping framework to track early interactions among winter wheat, FHB, and Trichoderma gamsii T6085. Two cultivars, Bingo and Rebelde, with higher and lower FHB susceptibility, respectively, were treated with a chemical fungicide (Chem) or T. gamsii T6085 (Bioc) under FHB pressure. Leaf- and canopy-level spectra were acquired at 2, 5, and 14 days post-inoculation, alongside gas exchange, water status, and chlorophyll measurements. Permutational multivariate analysis of variance (PERMANOVA) tested whole-spectrum effects, partial least squares discriminant analysis (PLS-DA) explored class separability, and partial least squares regression (PLSR) estimated physiological traits. PERMANOVA detected genotype × inoculation × treatment interactions from 5 days post-inoculation at leaf and canopy levels. PLS-DA revealed treatment- and cultivar-dependent spectral fingerprints, but overall low-to-fair validation performance indicates that these class-specific patterns should be interpreted as exploratory and not as evidence of operational treatment discrimination. PLSR provided high accuracy for chlorophyll content and osmotic potential, moderate accuracy for CO 2 -assimilation traits, and poor accuracy for transpiration and leaf water potential. While the workflow is scalable as an experimental and analytical framework, its operational deployment will require broader validation across sites, seasons, cultivars, disease-pressure conditions, and sensing platforms.
Detecting cotton leaf diseases in open-field environments is challenging due to cluttered backgrounds, scale variation, and irregular lesion morphology. Conventional detectors rely on isotropic receptive fields and coupled box-regression losses, which limit their ability to localize elongated lesions with poorly defined boundaries. We present an anisotropic boundary-aware detection framework that propagates high-frequency boundary information across four successive pipeline stages. In the backbone, an Anisotropic Morphological Contrast Aggregation module (AMCA) enhances direction-aware representation and lesion-background contrast via re-parameterizable strip convolutions and high-frequency residual extraction. A Dynamic Semantic Boundary Transfer mechanism (DSBT) then captures boundary priors from shallow layers before they are lost to downsampling and injects them into the neck. A Morphological-Spectral Synergistic Feature Pyramid Network (MFS-FPN) preserves these cues during multi-scale fusion through spatial-domain operations compatible with edge hardware. Finally, an Anisotropic Boundary-Decoupled IoU loss (ABD-IoU) independently penalizes each of the four box boundaries and sustains optimization signals in high-IoU regimes via a logarithmic modulation factor. On the self-constructed Complex Cotton Leaf Disease dataset (CCLD; 6,856 images, 6 classes), the method achieves 78.50% mAP@50 and 65.00% mAP@50:95, improving the YOLOv11n baseline by 4.80% and 2.70% with only 2.73 M parameters at 202 FPS. Cross-domain evaluations on PlantDoc and RWD confirm consistent improvements. The framework runs in real time on NVIDIA Jetson edge platforms with INT8 quantization.
Reproduction assets foundThe authors explicitly state that their source code, trained models, and implementation details are publicly available, and the data availability statement points to the same repository, which hosts the self-constructed CCLD cotton leaf disease dataset (6,856 images, 6 classes) used for the paper's phenotyping/disease-Code · publicFurthermore, to facilitate future research, our source code, trained models, and implementation details have been made publicly available at https://github.com/DynaVLA/ABAD-CLD .Open asset ↗DynaVLA/ABAD-CLDlines:331-343Dataset · publicThe datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://github.com/DynaVLA/ABAD-CLD .Open asset ↗DynaVLA/ABAD-CLDlines:1278-1317Plant phenotyping relevance matchEurope PMC · Crossref · checked 15 Sept 2026
Plant diseases lead to substantial yield losses and pose a persistent threat to global food security, creating an urgent demand for high-throughput, accurate, scalable, and non-destructive disease monitoring approaches. Remote sensing has emerged as a powerful tool, yet progress in plant disease detection remains fragmented across various disciplines, tasks, sensing methods, and data modalities. This review introduces a hex- view perspective to synthesise remote sensing–based plant disease detection within a cohesive conceptual framework. Instead of treating sensing technologies, algorithms, and datasets independently, the hex-view incorporates six interconnected dimensions that jointly capture how biological processes, measurement scale, and data characteristics constrain disease detectability, including when detection is possible and how reliably it can be achieved. The hex-view framework comprises six interconnected dimensions and forms an integrated framework called BTSCAD: (1) Biology (B): Plant-pathogen interactions constituting the biological foundation of disease development and expression. (2) Task (T): The diverse disease detection tasks and their corresponding research objectives. (3) Sensor (S): The sensing modalities that define the data acquisition type and richness of captured information. (4) Condition (C): The environmental conditions, sensing platforms, and spatial scales that shape disease observations and bridge controlled experiments and real-world deployment across leaf, canopy, plot and regional scales. (5) Algorithm (A): The classical and state-of-the-art data analysis algorithms used to extract disease-related information from sensor data. (6) Dataset (D): The data sources that underpin model development, evaluation, and generalisability. The hex-view perspective provides a clear framework for interpreting previous research and identifying future research directions. This review lays a structured foundation for developing robust, interpretable, and transferable disease detection systems, supporting advancements in precision agriculture, high-throughput phenotyping, and sustainable crop production.
Accurate identification of crop varieties is essential for plant breeding programs and the protection of Plant Breeders' Rights (PBR), yet traditional morphological assessment methods remain subjective and time-consuming, particularly for species with complex morphological diversity such as Rubus crataegifolius . This study demonstrates that geometric morphometric techniques provide an objective, quantitative complementary approach for distinguishing Korean raspberry varieties, addressing the limitations of subjective visual assessment while remaining compatible with molecular marker analysis. We employed three complementary morphometric approaches: landmark-based analysis (19 anatomical points capturing vein junctions and leaf margins), Elliptic Fourier Descriptors (EFD) for outline contours, and a hybrid landmark-EFD dataset. Using these approaches, we analyzed primocane and floricane leaves from 10 accessions of R. crataegifolius comprising 8 varieties and 2 landraces and performed principal component analysis (PCA) and linear discriminant analysis (LDA) with leave-one-out cross-validation. As a result, among the three morphometric approaches applied to primocane and floricane leaves, landmark-based analysis of primocane leaves achieved the highest classification accuracy (87.2%), with an overall average accuracy of 72.0% (range: 51.1-87.2%) across all six analytical combinations. LDA visualization suggested the presence of four major morphological groups, and primocane leaves exhibited higher discriminatory power than floricane leaves, which may reflect greater morphological uniformity under normal growing conditions. Landmark analysis effectively detected subtle differences in leaf venation and leaflet architecture that are difficult to distinguish visually, highlighting the capacity of morphometrics for objective and multidimensional morphological analysis. These findings suggest that morphometric analysis provides a practical and cost-effective preliminary screening tool, complementary to molecular approaches, for supporting Distinctness, Uniformity, and Stability (DUS) examination in raspberry variety evaluation. This approach shows strong potential for offering a scalable solution for variety registration and protection and supporting sustainable horticultural development.
Assessment of chlorophyll content is important to understand plant nitrogen status in precision agriculture. Traditional destructive methods for chlorophyll quantification are time-consuming, labor-intensive, and unsuitable for high-throughput phenotyping applications. The SPAD meter (Soil Plant Analysis Development) provides a rapid and non-destructive alternative by measuring leaf greenness as a proxy for chlorophyll content. Recent technological advances in imaging sensors and computational methods have enabled the development of low-cost approaches for predicting SPAD values. In this study, we propose an ensemble deep learning model-based Android application ( SPAD Predictor ) that was developed for predicting the SPAD value from RGB contact imaging. A total of 34 features, including color space features, RGB-derived features, and vegetation indices, were used to develop the model. The model consists of a lightweight Multi-Layer Perceptron (MLP) and Random Forest (RF) Regressor layer with stacking ensemble architecture. A linear regression was used as a meta-model to ensemble the MLP and RF layers. A permutation-based feature importance analysis showed that the a* channel, ExGR, RG, NRI and VARI indices played the most important roles in predicting SPAD value. The proposed ensemble deep learning model yielded R 2 t r a i n i n g of 0.987 and R 2 t e s t i n g of 0.89, RMSE of 3.25. The developed application was successfully deployed and was able to perform image submission, backend communication, prediction generation, result display, and history management. Field-level validation of the developed application yielded R 2 of 0.848, RMSE of 3.068, and MAE of 2.544. These findings indicate that the developed system has practical potential as a low-cost, field-applicable tool for estimating paddy leaf SPAD.
India’s economy is primarily based on agriculture. Agriculture has significant contribution in nation’s GDP. Food security and employment significantly influenced by agriculture. However factors like uncertain weather conditions, poor quality of seeds and plant diseases impact on agriculture productivity. Computer vision and DL algorithms are most crucial components of precision agriculture. Early detection can improve decision making, maximize pesticide use, and preserve harvests. Using CNN architectures, segmentation-based approaches, handcrafted feature-based methods, and hybrid approaches incorporating Machine Learning and Deep Learning this study seek to provide review of recent publications from 2020 to 2026. The review was carried out using a variety of publications with different datasets, methodologies, and outcomes. The findings show that DL, especially CNN and transfer learning models, performed better than machine learning techniques. It points out several significant problems, such as dataset imbalance, insufficient generalization, computing inefficiency, and a dearth of real-world data. Future research topics are also suggested which includes IoT-driven real-time solutions, lightweight architecture, domain adaption, and multimodal imaging. This review aims to develop plant disease detection technologies that are more dependable, scalable, and field deployable.
To address this challenge, we propose StyleGAN3-T, the translation-equivariant alias-free variant of StyleGAN3, as a generative framework for producing high-fidelity synthetic plant disease images, integrated with a hybrid Swin Transformer-ResNet50 classifier for precise recognition. Accurate detection of plant leaf diseases is essential for sustainable agriculture and early intervention. However, deep learning models often struggle with small, imbalanced datasets that limit generalization and robustness. To address this challenge, we propose StyleGAN3-T, a novel alias-free generative framework for producing high-fidelity synthetic plant disease images, integrated with a hybrid Swin Transformer-ResNet50 classifier for precise recognition. The proposed approach ensures translation-equivariant, artifact-free image synthesis and enhanced feature diversity. A balanced dataset of 18,000 images was developed by combining real and StyleGAN3-T-generated samples. In pooled GAN benchmarking, StyleGAN2-ADA achieved the strongest generative-quality metrics, whereas StyleGAN3-T was selected as the preferred augmentation model because its alias-free synthesis and spatial consistency yielded superior downstream classification performance in the proposed pipeline.
Reproduction assets foundThe paper's grape leaf disease image inputs are two publicly available Kaggle datasets explicitly named in the Data Availability statement. No author code, models, or synthetic dataset deposit is provided; other processed data is request-only.Dataset · publictechnical guidance. Y.L. and A.W. supervised the study, provided critical revisions, and contributed to the interpretation of results. All authors reviewed and approved the final manuscript.
Data availability
The datasets analyzed during the current study are publicly available from Kaggle: Grapevine Disease Dataset (Original) (https://www.kaggle.com/datasets/rm1000/grape-disease-dataset-original; accessed 13 March 2026; license: MIT) and Grape Leaf Disease 4 Class (https://www.kaggle.com/datasets/jawadulkarim117/grape-leaf-disease-4-class; accessed 13 March 2026; license: CC0: Public Domain). Additional processed metadata, label-harmonization records, dataset split definitions, and other daOpen asset ↗Kaggle · rm1000/grape-disease-dataset-originallines:549-576Dataset · publicthors reviewed and approved the final manuscript.
Data availability
The datasets analyzed during the current study are publicly available from Kaggle: Grapevine Disease Dataset (Original) (https://www.kaggle.com/datasets/rm1000/grape-disease-dataset-original; accessed 13 March 2026; license: MIT) and Grape Leaf Disease 4 Class (https://www.kaggle.com/datasets/jawadulkarim117/grape-leaf-disease-4-class; accessed 13 March 2026; license: CC0: Public Domain). Additional processed metadata, label-harmonization records, dataset split definitions, and other data used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Declarations
Competing inOpen asset ↗Kaggle · jawadulkarim117/grape-leaf-disease-4-classlines:549-576Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
Film mulching can promote maize canopy development by altering soil thermal conditions. However, commonly used air-temperature-based growing degree days (GDDs air ) may not adequately reflect mulch-induced soil warming or the effects of biodegradable film degradation on leaf area index (LAI) dynamics. To improve unified simulation of maize LAI under different film mulching conditions, field experiments were conducted in 2023 and 2024. Five treatments were established: 0.006, 0.008 and 0.010 mm biodegradable films (DM1, DM2 and DM3, respectively), a 0.010 mm conventional plastic film (PM), and a no-mulching control (CK). The compensation of increased soil temperature for air-temperature-based thermal accumulation during early maize growth was quantified. Modified Logistic LAI models were then developed using days after emergence (DAEs), GDDs air , soil-temperature-compensated growing degree days (GDDs stc ), and normalized GDDs stc (NGDDs stc ) as driving variables. The models were calibrated with observations from 2023 and independently validated with observations from 2024. The compensation effect acted through mulch-induced increases in 0-10 cm soil temperature during early maize growth and was stronger at the seedling stage than at the jointing stage. Compared with DM1 and DM2, daily compensation values were higher by 0.25-0.78 °C under DM3 and by 0.26-0.76 °C under PM. Independent validation showed that the GDDs stc -driven model had lower prediction error than the DAEs- and GDDs air -driven models. The NGDDs stc -driven model performed best; its RMSE values were 17.61%, 15.17% and 10.91% lower than those of the DAEs-, GDDs air - and GDDs stc -driven models, respectively. These results indicate that incorporating mulch-induced soil temperature compensation into the thermal time scale can more accurately represent maize canopy development under film mulching conditions.
The genetic identity of coffee cultivars is fundamental to the specialty coffee sector, where premium prices are paid under the assumption that the purchased planting material corresponds to the declared variety. However, many producing countries lack the certification infrastructure necessary to guarantee this identity in their informal seed systems, exposing producers to undetected varietal non-conformity. In this study, we examine a case from a specialty coffee ( Coffea arabica L.) farm in southern Ecuador where seeds labeled as Sidra (USD 100/kg) and Gesha (USD 500/kg) were purchased without genetic or phytosanitary certification. Using a combination of SSR-based DNA fingerprinting and quantitative morphological characterization, including plant architecture, leaf functional traits, and fruit characteristics, we documented varietal identity and assessed the discriminant capacity of morphological traits across the four resulting morphotypes. Using eleven microsatellite markers for SSR fingerprinting, we found that two of the four morphotypes did not match their declared commercial identity. One plant sold as Sidra was identified as compatible with Batian, a composite variety of Kenyan origin that is genetically unrelated to Ethiopian landraces. The plants acquired as Gesha corresponded to a pure Ethiopian landrace that is genetically similar to, but not identical to, the Panamanian Geisha reference accession T.02722. Only two morphotypes were confirmed as Sidra. Furthermore, the placement of Sidra within the Core Ethiopia genetic group is consistent with prior population-level analyses and with its likely status as a selected Ethiopian landrace rather than a variety of hybrid origin. Morphological linear discriminant analysis achieved 82.4% overall classification accuracy under leave-one-out cross-validation (LOOCV), with internode length dominating the first discriminant function (LD1 = 66.6%). These results demonstrate that varietal nonconformity in the specialty coffee seed sector can extend to the inadvertent introduction of genetically unrelated material and underscore the urgent need for accessible seed certification.
Reproduction assets foundThe paper's morphological/functional trait dataset (used for the phenotyping and LDA analysis) is explicitly stated to be publicly available on Figshare (10.6084/m9.figshare.32841344). No author analysis code repository is stated; other URLs in the text are generic libraries or cited prior work.Dataset · publicThe morphological and functional trait dataset generated and analyzed in this study is publicly available in the Figshare repository at 10.6084/m9.figshare.32841344 .Open asset ↗Figshare · 10.6084/m9.figshare.32841344lines:526-568Plant phenotyping relevance matchOpenAlex · Crossref · Europe PMC · checked 5 Sept 2026
Here, we present a single-operator push-cart platform equipped with a 16-beam LiDAR. A push-button interface controls data acquisition, and the data processing pipeline removes ground points, filters noise, performs 5-cm voxelization, and produces plot-level canopy metrics. We validated biomass estimation in hairy vetch (Vicia villosa) and corn (Zea mays) leaf- and whole-plant thinning experiments. In vetch, voxelized estimation of plant volume correlated strongly with destructively measured biomass (r2 = 0.88), showing that the multi-beam LiDAR can produce biomass estimates comparable to previously reported methods. In corn, comparisons of perpendicular (0°) and multi-angle LiDAR beams showed significantly greater voxel counts in the upper canopy when angled beams were used (beam angle × height interaction, p < 0.001), demonstrating that multi-beam scanning provides greater penetration into the upper canopy than a single perpendicular scan plane. We also extended the suite of LiDAR-derived traits to include apparent leaf area index (LAI), mean tilt angle (MTA), persistent homology-based stand density, and plot-bounded foliage area density (FAD). The persistent homology algorithm distinguished between leaf-removal and plant-removal treatments (removal type × removal amount, p = 0.0039). LiDAR-derived LAI has been used to estimate canopy leaf area, but gap-fraction approaches do not fully exploit the ability of LiDAR to resolve distance. Plot-bounded FAD used ray length and interception distance within defined plot volumes and was more sensitive to plot-level treatments than apparent LAI or MTA, detecting differences associated with both the removal amount and removal type. These results show that a robust, portable, multi-beam LiDAR cart can reproduce plot-level canopy measurements and improve trait especially in research-sized plots.
Abstract Crop disease is a major worldwide problem in agricultural production and food security, adversely affecting yield and quality for a variety of plant species. To overcome these drawbacks, this research provides a Double Transfer Learning-based Capsule Network (DTL-CapsNet) approach for automated plant disease classification with multiple crops. Based on the image pre-processing, segmentation, double transfer learning, and Capsule Networks technologies, the proposed framework extracts discriminative features of the diseases and maintains spatial relations between the leaves symptoms effectively. This double transfer learning approach involves extracting general visual features from pre-trained deep learning models and then fine-tuning these features to classify plant diseases. Capsule Networks then leverage the visual similarity of disease patterns to make the recognition more robust, while simultaneously adding hierarchical part–whole relationships in leaf structures, thereby improving the feature representation. Experiments were performed on a heterogeneous data set consisting of 21,927 leaf images belonging to 17 different healthy and diseased classes of apple, chilli, cotton, corn and potato crops. The experimental results proposed DTL-CapsNet framework is more accurate compared to the traditional CNN-based models and conventional transfer learning models. The proposed method of double transfer learning and Capsule Networks offers an efficient and scalable approach for intelligent plant disease diagnosis, offering significant potential in precision agriculture and real-time crop monitoring systems.
Early detection of diseases in plants has been identified as a critical factor for ensuring the maintenance of productivity, preventing economic losses, and promoting sustainable agriculture. Traditional manual approaches for diagnosing diseases are time-consuming, subjective, and inappropriate for large-scale and real-time agriculture. In order to overcome the limitations of traditional approaches, the CNN–CBAM–MobileViTNet has been proposed, an efficient attention-guided network by the fusion of Convolutional Neural Networks (CNNs), Convolutional Block Attention Module (CBAM), and Mobile Vision Transformer (MobileViT) for plant diseases recognition.The CNN component is effective in capturing local visual patterns like lesions, discoloration, and texture. The CBAM component is effective in refining the feature representations by focusing on disease-related spatial areas and useful channels. The MobileViTNet branch is useful in capturing contextual relationships from the leaf areas through lightweight transformer blocks. The CNN–CBAM—MobileViTNet is tested on an enhanced dataset with 38 classes of plant diseases and health conditions, splitting data into 70% training, 15% validation, and 15% testing. Significantly, extensive experimental analysis reveals that the test accuracy is 99%, with high precision, recall, and [Formula: see text]1-score values. Training-validation curves show that the model converges stably with little overfitting, while ROC analysis shows high classwise discrimination ability of the model. Hence, the CNN–CBAM–MobileViTNet model is reliable and may be used for real-world applications in smart agriculture and automatic crop disease monitoring systems.
Abstract Precision field management and high-throughput plant phenotyping increasingly rely on remote sensing to capture spatial and temporal variability in crop performance. Unmanned aerial vehicle (UAV) – based sensing offers unique advantages for field-scale data collection, including high spatial resolution, flexible deployment, and scalable throughput. However, the full potential of UAV platforms remains constrained by labor-intensive operations across flight execution, data transfer, and processing workflows. This study presents a systematic evaluation of an automatic UAV-based crop sensing platform through a season-long, multi-crop field experiment. Data acquisition was conducted over a maize irrigation trial and a soybean breeding experiment, resulting in 176 completed flights over 28 days during the growing season. High-frequency flights on selected days captured diurnal dynamics in key canopy traits, including maize leaf rolling under drought stress and genotype-dependent plot temperature variation in soybean. In the maize irrigation experiment, significant differences in diurnal canopy cover ratio (CCR) were observed among irrigation treatments. The predictive relationship between CCR and final grain yield strengthened throughout the day, with the coefficient of determination (R 2 ) increasing from 0.05 in the early morning (RMSE = 3.05 Mg ha − 1 ) to 0.65 at midday (RMSE = 1.87 Mg ha − 1 ), highlighting the importance of temporal optimization in UAV-based sensing. Temperature measurements from the onboard thermal infrared camera showed a strong overall linear correlation with ground truth measurements (R 2 = 0.85). In the soybean trial, the highest plot temperature was observed on the fast-wilting genotype. Additionally, regression models were developed to estimate key crop traits, including canopy height (CH) and leaf area index (LAI), demonstrating the platform’s quantitative sensing capability. Overall, this study demonstrates that automatic UAV systems enable high-temporal-resolution crop monitoring while substantially reducing operational cost. The results highlight their potential for precise crop management and scalable field phenotyping. Future work will focus on integrating automated data processing pipelines to support near-real-time analytics and decision-making.
ABSTRACT While whole-genome sequencing captures millions of single nucleotide polymorphisms (SNPs) and hyperspectral imaging (HSI) enables non-destructive plant phenotyping, integrating these modalities to link genotype to phenotype remains challenging due to their high dimensionality and non-linearity. This study presents DeepPheno a deep learning framework that predicts SNP genotypes from HSI data, using model predictability as a proxy for genotype-phenotype association. HSI data were acquired from 194 lettuce genotypes under field conditions. HSI data patches (20×20 pixels × 224 spectral bands) were used to train a hybrid CNN to predict the variant of a specific SNP. The framework was validated on SNPs with known phenotypic effects (anthocyanin, leaf serration, pale pigmentation), achieving high predictive performance (AUC ranging from 0.806 to 0.935), whereas models trained on randomly shuffled labels performed at chance (mean AUC ≈ 0.51). Extending the workflow to 50 randomly selected putatively neutral SNPs, most yielded low predictability, but two showed high performance (AUC > 0.76), suggesting uncharacterized genotype-phenotype links. Explainable AI, including SHAP and Grad-CAM, identified relevant spectral and spatial features driving these predictions, particularly the green and red-edge wavelengths associated with pigment dynamics and leaf structure. These results establish a framework for understanding complex genotype-phenotype interactions in plants and extracting these links from HSI data without predefining the exact trait values. It provides an avenue for high-throughput trait discovery and description and extends the integration of image-based phenomics with plant genetics.
Reproduction assets foundThe paper's Data Availability statement deposits authors' code, scripts, and supplementary material in a public GitHub repository, including a downscaled de-identified sample dataset demonstrating the pipeline. The raw HSI/genotype datasets are proprietary under NDA and not public.Code · publicThe code, scripts, and supplementary material supporting the findings of this study have been deposited in the GitHub repository at https://github.com/frankgyan/Utrecht-University--HSI .Open asset ↗frankgyan/Utrecht-University--HSIlines:195-223Code / dataset availability confirmedCrossref · checked 14 Sept 2026
Plant diseases threaten global agriculture, causing 20–40% yield losses and food insecurity. Current diagnostic methods are costly and lack scalability. While deep learning advances plant disease detection, there remains a need for CNNs with simpler architectures, better generalizability, and lower computational cost. This study presents a novel CNN for multi-class classification of 38 diseases. Trained on a public dataset of over 87,000 RGB images, the architecture comprises five convolutional blocks (filters 32–512) with max pooling and dropout (0.25, 0.4), followed by a 1,500-unit dense layer and SoftMax output. Optimized with Adam (lr=0.0001) and categorical cross-entropy, the model achieved 98% training and 96% validation accuracy with approximately 28.7 million parameters significantly fewer than transfer learning architectures. These results demonstrate an effective balance between predictive performance and computational efficiency, positioning the model as a promising tool for real-world agricultural deployment.
Reproduction assets foundThe paper's sole qualifying asset is the plant disease image dataset used for all its CNN training/validation measurements: the publicly available New Plant Diseases Dataset (Augmented) on Kaggle, explicitly declared in the Data availability statement. No author code, trained model checkpoints, or other paper-specific Dataset · publict to disclose.
Acknowledgment:
The authors are sincerely grateful to their institutions for their continued support and trust, which greatly contributed to
the completion of this research.
Data availability
The dataset used and analyzed during the current study, “New Plant Diseases Dataset”, is publicly available on Kaggle at:
(https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset?select=New+Plant+Diseases+Dataset%28Augmented%29)
References
[1] P. Arputharaj and K. Karunanithy, "A review on machine learning and deep learning techniques for plant leaf disease
detection and classification with IoT in agriculture industry," Journal of Industrial Information Integration, vol. 50, Open asset ↗Kaggle · vipoooool/new-plant-diseases-datasetpdf-raw-page:11 lines:1-49Dataset · publict to disclose.
Acknowledgment:
The authors are sincerely grateful to their institutions for their continued support and trust, which greatly contributed to
the completion of this research.
Data availability
The dataset used and analyzed during the current study, “New Plant Diseases Dataset”, is publicly available on Kaggle at:
(https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset?select=New+Plant+Diseases+Dataset%28Augmented%29)
References
[1] P. Arputharaj and K. Karunanithy, "A review on machine learning and deep learning techniques for plant leaf disease
detection and classification with IoT in agriculture industry," Journal of Industrial Information Integration, vol. 50, Open asset ↗Kaggle · vipoooool/new-plant-diseases-datasetpdf-raw-page:11 lines:1-49Code / dataset availability confirmedEurope PMC · bioRxiv · checked 15 Sept 2026
Accurate segmentation of plant disease symptoms is essential for crop monitoring and phenotyping, yet it typically requires costly pixel-level annotations. Weakly supervised semantic segmentation (WSSS) alleviates this burden using image-level labels, but its performance depends on the quality of spatial priors such as class activation maps (CAMs). We investigate whether text-guided segmentation with the Segment Anything Model 3 (SAM3) can serve as an alternative weak supervision signal. Three pseudo-mask generation strategies are compared: (i) CAMs refined with SAM or SAM3, (ii) zero-shot text-guided SAM3, and (iii) a hybrid approach combining weak spatial cues with text prompts. The resulting pseudo-masks are used to train a DeepLabV3 model. Text guidance alone matches or outperforms conventional WSSS, achieving up to 0.46 IoU without spatial supervision and 0.61 IoU on a public dataset, although performance is sensitive to text prompt formulation. The hybrid strategy improves robustness, reaching 0.50 IoU on the primary dataset and 0.58 IoU on the additional dataset while reducing prompt sensitivity. Overall, text guidance is a promising alternative to conventional weak supervision, while hybrid approaches provide a more robust solution for plant disease segmentation.
Reproduction assets foundThe paper's own oilseed rape leaf disease image dataset (2,540 RGB images, 601 pixel-level annotations) is explicitly made publicly available on Recherche Data Gouv under DOI 10.57745/R3HZOQ, which is an allowed URL. A companion dataset describing the foliar pathogens (DOI 10.57745/UCEKI8) is also public. The pixel-annDataset · publicAn original dataset of 2,540 RGB images of oilseed rape leaves affected by
7 fungal and bacterial diseases (Table 1) was built to develop and evaluate
the proposed pipelines, and is made publicly available [25].Open asset ↗pdf-raw-page:5 lines:1-43Plant phenotyping relevance matchEurope PMC · checked 5 Sept 2026
Introduction Maize is one of the most important food crops in the world, and foliar diseases can lead to significant yield losses if identification is not performed on time. Experts conducting manual inspections find it less effective and more subjective. Deep learning-based approaches utilizing Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have been demonstrated as a viable approach to automate disease diagnosis. Thus, while CNNs fail to capture wider context due to their local feature focus and ViTs need larger datasets and tend to miss finer-grained details. To overcome these limitations, we present EDISP a hybrid framework that connects Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for local feature extraction, as well as global contextual learning. Methods The EDISP framework brings together the strengths of CNNs and ViTs to overcome their individual weaknesses. It is trained on a dataset that includes both controlled-environment and real-field maize leaf images, which helps it handle different environmental conditions. The data undergoes thorough preprocessing, including normalization, augmentation, and stratified splitting into training, validation, and test sets to support generalization. The CNN focuses on detailed local disease features, while the ViT captures broader contextual information across the maize leaf surfaces. Results The proposed EDISP model significantly outperforms standalone CNN and ViT Models in multiple performance metrics, achieving an overall classification accuracy of 99.40%, precision of 99.43%, recall of 99.38%, and an F1-score of 99.40%. Experimental results demonstrate that EDISP excels in identifying maize leaf diseases, including Common Rust, Gray Leaf Spot, Northern Leaf Blight, and Healthy leaves, with minimal false positives and negatives. External validation with an independent dataset further highlights the model's robustness and ability to generalize to real-world conditions. Discussion The EDISP hybrid architecture, integrating CNNs and ViTs, provides a stronger method for accurate, automated maize leaf disease detection. Its robust performance, consistent results on controlled and field datasets shows robustness in diverse environments. However, EDISP's effectiveness may be limited by image quality, lighting, or disease types not seen in training. These results highlight the promise of hybrid deep learning in precision agriculture and offer a scalable solution for disease detection, supporting farmers without expert diagnostic resources.
Based on fluorescence hyperspectral imaging (FHSI), this study targeted rapid, non-destructive quantification of lead (Pb) content in oilseed rape leaves treated with varying silicon (Si) concentrations, acquiring fluorescence spectra over the 484.43-1001.61 nm wavelength range. To optimize spectral data quality, preprocessing methods (Savitzky-Golay smoothing, first derivative, detrending) were comprehensively compared. Characteristic wavelengths were then selected via interval variable iterative shrinkage, which effectively compressed data dimensionality and reduced computational load. A hybrid SE-CL1DA model, fusing a 1D convolutional neural network, a long short-term memory network and SE attention mechanism was constructed, with Bayesian optimization tuning hyperparameters to boost stability. The BO-SE-CL1DA outperformed both traditional machine learning and insufficiently optimized deep learning model (R p 2 =0.9609, RMSE = 0.0377 mg/kg, RPD = 5.1736), thus enabling accurate Pb estimation, supporting Si-regulated heavy metal stress management and facilitating agricultural contamination monitoring.
Marcel M. El Hajj · Kasper Johansen · Oliver M. Lopez Valencia · Fabio Veiga de Camargo · Yu-Hsuan Tu · Samer K. Al Mashaharawi · Omar A. López Camargo · Victor Angulo · Dominique Courault · Matthew F. McCabe
Purpose In recent years, there has been a growing use of unmanned aerial vehicle (UAV) based light detection and ranging (LiDAR) data for mapping plant area index (PAI) in orchards. However, using LiDAR time-series collected throughout the growing season to assess PAI variations in response to phenology, represents an understudied area of investigation. Furthermore, establishing the optimal spatial resolution for mapping biophysical variables of tree crops from LiDAR point cloud data remains poorly defined. Here, we assess the capability of a UAV-based LiDAR system to characterize cherry trees throughout the growing season, with a focus on monitoring PAI and the vertical structure of individual trees. Methods A time-series of 14 point cloud acquisitions with a density of 3300 points/m2 was collected between February and December 2022, covering all phenological stages of a cherry orchard in southern France. A voxel-based method was applied to create a three-dimensional grid within which PAI was estimated for each voxel. PAI was mapped by accumulating the individual voxel-based PAI values within each vertical voxel column. Results The results demonstrate that a voxel size of at least 0.7 m is required to retrieve reliable PAI estimates (RMSE = 0.58 m2.m−2, MAE = 0.48 m2.m−2, bias = 0.19 m2.m−2, rRMSE = 23%, and R2 = 0.51), while a voxel size of 1 m produced the most accurate PAI estimates (RMSE = 0.5 m2.m−2, MAE = 0.41 m2.m−2, bias = 0.07 m2.m−2, R2 = 0.59), when assessed against field-based PAI measurements obtained with a LAI-2200 Plant Canopy Analyzer. The temporal variation of canopy PAI illustrated the progression of key phenological stages, including flowering, leaf development, ripening and senescence, as well as the response of the canopy to drought stress (reduction in PAI due to leaf rolling) during the summer. The maps of PAI successfully described the variations in leaf canopy density for different cherry varieties and allowed assessment of the vertical PAI profile at the individual tree level, which provides valuable insight into tree condition. Conclusion This study confirms that seasonal UAV-LiDAR monitoring is a viable, informative approach for capturing orchard canopy dynamics at the individual tree and sub-canopy level, linking canopy structure to phenology, varietal differences, and stress responses across the growing season.
Agricultural leaf disease detection is crucial for early intervention and yield protection in precision agriculture. Among representative economic crops, such as apples, leaf lesions are typically small and appear in complex backgrounds, making accurate detection performed on resource-constrained embedded devices challenging. To address this, we propose a lightweight small-object detection models, namely the dynamic Differential Compensation Lightweight-YOLO (DCL-YOLO) model and its pruned version (DCL-YOLO-P), based on YOLO11n. A novel Dual-Aspect Feature Complementary Mapping (DAFCM) module type is embedded in their backbone to recover lost semantic and spatial information, while the original YOLO11n's neck is replaced by an Efficient Enhanced Cross-Scale Feature Fusion (EE-CSFF) module, which incorporates Gated Differential Convolutional Fusion (GDCF) modules to strengthen cross-scale information flow and small-object representation. Experimental results obtained on the ALDSOD dataset show that, compared with the YOLO11n baseline, DCL-YOLO improves recall from 81.9% to 84.6%, mAP50 from 86.8% to 88.4%, and mAP50:95 from 47.0% to 47.8%, while also reducing the parameter count from 2.58 M to 1.91 M and Giga Floating-Point Operations (GFLOPs) from 6.3 to 5.5. After applying Layer-Adaptive Magnitude-based Pruning (LAMP), the parameter count and GFLOPs are further reduced to 0.75 M and 2.7, respectively, with mAP50 and mAP50:95 still exceeding the baseline by 1.2 and 0.5 percentage points, respectively. When deployed on an embedded device, the pruned model achieved 15.2 FPS and 139 msec per image, confirming its applicability in real-time scenarios. Furthermore, cross-domain validation, performed on the Global Wheat Head Detection (GWHD) dataset, indicates the stable generalization capabilities of the proposed models across environmental domain shifts. The DCL-YOLO's source code is publicly available at: https://github.com/q123-code/dcl-yolo.
Reproduction assets foundThe paper's constructed ALDSOD apple-leaf disease detection dataset is publicly available via Zenodo DOI, and the authors' DCL-YOLO source code is publicly available on GitHub. Both are paper-specific, public, and actionable.Dataset · publicData Availability: The constructed ALDSOD dataset used in this study is available for download from the following DOI: https://doi.org/10.5281/zenodo.17198053 .Open asset ↗zenodo · 10.5281/zenodo.17198053lines:148-159Code · publicThe DCL-YOLO’s source code is publicly available at: https://github.com/q123-code/dcl-yolo .Open asset ↗github · q123-code/dcl-yololines:148-159Code / dataset availability confirmedEurope PMC · Crossref · checked 5 Sept 2026
Abstract Plant diseases affecting leaves and fruits cause substantial yield and economic losses worldwide, particularly in horticultural crops cultivated under diverse agro-climatic conditions. Early and accurate disease diagnosis is essential for effective crop management; however, manual inspection is time-consuming, subjective, and often infeasible at large scale. In this work, we present the Tomato–Chilli–Papaya (TCP) Fruit and Leaf Disease Dataset, a comprehensive multi-crop image dataset designed to support deep learning-based plant disease recognition. The dataset comprises labeled RGB images of healthy and diseased leaves and fruits from three economically important crops—tomato, chilli, and papaya—captured under real-field and semi-controlled environments, reflecting significant variability in illumination, background complexity, and disease severity. To demonstrate the applicability of the dataset, several commonly used convolutional neural network (CNN) architectures, including VGG, ResNet, DenseNet, MobileNet, and EfficientNet models, were trained and evaluated on the TCP dataset using transfer learning. Experimental results show that deep CNN models can effectively learn discriminative visual features corresponding to disease-specific patterns such as leaf spots, lesions, discoloration, curling, and fruit surface abnormalities. Lightweight models such as MobileNet achieve competitive performance with reduced computational cost, while deeper architectures provide improved accuracy at the expense of higher complexity. The results highlight the importance of dataset diversity for robust model generalization across multiple crops and plant organs. The TCP dataset provides a challenging benchmark for single-crop and multi-crop disease classification and supports the development of advanced deep learning, attention-based, and explainable AI models for precision agriculture. By enabling reproducible research and realistic performance evaluation, this dataset contributes toward scalable and practical AI-driven plant disease diagnosis systems aimed at reducing yield losses and supporting sustainable agriculture.
Reproduction assets foundThe paper introduces the TCP (Tomato-Chilli-Papaya) fruit and leaf disease image dataset and reports CNN experiments on it. The dataset is publicly deposited on Mendeley Data, and the authors state that analysis code is available on GitHub. Both are paper-specific, public, and actionable.Dataset · publicData is available on Mendeley:1Open asset ↗pdf-page:27 lines:1-51Code · publicCode availability: Code is available on GitHub 2Open asset ↗GitHubpdf-page:27 lines:1-51Plant phenotyping relevance matchOpenAlex · Crossref · Europe PMC · checked 5 Sept 2026
Introduction: Plant phenotyping requires accurate and repeatable three-dimensional structural information, but practical acquisition conditions in greenhouses, seedling rooms, and indoor pot experiments often include complex backgrounds, handheld motion blur, and thin leaf structures. These factors reduce the robustness of conventional three-dimensional reconstruction methods and limit their use in low-cost and automated phenotyping. Methods: To address this problem, this paper proposes F2DMAS, an automated three-dimensional plant phenotyping workflow using consumer-grade smartphone videos. The workflow first converts multiview RGB videos into image sequences and removes motion-blurred frames through frequency-domain quality filtering. A frequency-spatial plant segmentation module, termed FSAM3, is then introduced to separate plant structures from complex backgrounds without task-specific annotated training data. The segmented image sequences are further reconstructed using 2D Gaussian Splatting, followed by TSDF-based meshing, scale recovery, and virtual measurement for extracting plant height, canopy width, leaf length, and leaf width. Results: Experiments were conducted on 15 plant species under two acquisition scenarios. The proposed workflow achieved stable plant reconstruction under non-ideal background conditions, with PSNR, SSIM, and LPIPS values of 31.09, 0.9711, and 0.0365, respectively. Compared with the baseline reconstruction workflow, F2DMAS substantially reduced the processing time for mesh extraction while improving reconstruction quality. The extracted phenotypic traits showed strong agreement with manual measurements, with R² values ranging from 0.90 to 0.99, RMSE values ranging from 0.64 to 1.21 cm, and MAPE values ranging from 4.50% to 9.73%. Discussion: These results indicate that F2DMAS can provide an end-to-end workflow from smartphone video acquisition and plant segmentation to three-dimensional reconstruction and phenotypic trait extraction. The proposed method offers a practical and deployable solution for greenhouse seedling cultivation, potted plant experiments, and low-cost three-dimensional plant phenotyping.
Accurate lesion segmentation is essential for automated plant disease analysis in precision agriculture. Although the Segment Anything Model (SAM) exhibits strong generalization ability, its direct application to plant disease images in natural field environments remains challenging due to cluttered backgrounds, dense leaf veins, uneven illumination, and frequent occlusions. In particular, SAM mainly relies on global structural cues and is often insufficiently sensitive to subtle lesion textures and weak local details, which can result in missed small or early-stage lesions and inaccurate boundary delineation. To address these limitations, we enhance SAM with a disease-specific detail compensation module for plant disease lesion segmentation. A ResNet-50based branch is employed to extract fine-grained local texture features that are difficult for SAM to capture. These fine-grained features are fused with SAM encoder representations and then injected into the SAM decoder, enabling more accurate lesion prediction while preserving SAM’s strong global modeling capability. More importantly, we propose a reliability-guided variational fusion framework to further improve the interaction between heterogeneous features. Specifically, instead of conventional similarity or addition-based fusion, we introduce an uncertainty-aware variational fusion strategy that explicitly quantifies the confidence of each feature stream. An uncertainty encoder models feature distributions probabilistically, and a variational fusion module dynamically assigns higher weights to more reliable features while suppressing uncertain or interfering responses. In addition, Kullback-Leibler divergence regularization is introduced to stabilize cross-feature alignment and improve fusion robustness. Extensive experiments on PlantSeg, PlantDoc-Seg, and ATLDSD demonstrate that the proposed method outperforms state-of-theart approaches, achieving DSC scores of 81.05%, 91.12%, and 88.27%, respectively. The proposed method addresses SAM’s weakness in fine-grained disease feature extraction, accurately identifies early and small lesions, and delivers reliable segmentation for field plant disease automatic diagnosis.
Reproduction assets foundThe paper evaluates ReLeaf-SAM on three public plant disease segmentation datasets. One of them, PlantDoc-Seg, is explicitly a community-provided Kaggle dataset with a verbatim URL matching an allowed URL; it is a public plant image/mask dataset directly used for this paper's segmentation measurements. PlantSeg and ATLDataset · publicTherefore, we used a community-provided segmentation subset from Kaggle 1 , which we refer to as PlantDoc-Seg in this study. This subset is derived from PlantDoc and contains 588 diseased leaf images with corresponding binary masks, enabling supervised leaf disease segmentation.Open asset ↗Kagglelines:48-115Code / dataset availability confirmedEurope PMC · checked 5 Sept 2026
Agriculture is significant in world food production and global economic stability, but leaf disease and pest infection can cause a threat to crop quantity and quality. Thus, it became crucial to have timely and accurate identification of plant leaf disease to prevent loss in agriculture on a large scale and to have sustainable crop management. This paper introduces LeafLiteX, a lightweight mobile-based deep learning application for real-time detection and classification of leaf diseases. This application uses U-Net segmentation to precisely find leaf regions and MobileNetV3-Large to quickly classify diseases with less computation on the computer. The application performs end-to-end processing, from image acquisition to segmentation and disease prediction on mobile devices. An experiment was performed on publicly available crop disease datasets containing various leaf images having different disease types. The model obtained an accuracy of 98.85% showing improved generalization with minimal latency. The design of the model was such that it was suitable for inference on-device while still being robust enough despite changes in lighting conditions, background noise, and camera resolution. LeafLiteX is a low-cost, easy to use, offline-capable, and in-the-moment decision-making supportive diagnostic application that supports farmers and agrarians who require early detection. This paper demonstrates the capabilities that can be achieved using edge-optimized machine learning and computer vision to support the development of smart agriculture technologies. While traditional methods rely solely on classification, this research focuses more on practical implementation by incorporating segmentation, lightweight classification, and explainability to develop a mobile-friendly model.
Reproduction assets foundThe paper's Data availability statement explicitly links the public PlantVillage (Mendeley) and PlantDoc (GitHub) leaf-image datasets used for its experiments, and provides the authors' LeafLiteX source code on GitHub.Dataset · publicThe dataset used in this study is publicly available from the repository:
https://data.mendeley.com/datasets/tywbtsjrjv/1Open asset ↗data.mendeley.com · tywbtsjrjv/1pdf-page:29 lines:1-74Code · publicThe source code is available on the following link: https://github.com/phdpawan/LeafLiteX.Open asset ↗github.com/phdpawan/LeafLiteXpdf-page:29 lines:1-74Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Heat stress limits plant productivity by disrupting transpiration, altering leaf microclimate, and activating metabolic pathways that increase volatile organic compound (VOC) emissions. VOC signatures, combined with leaf-level relative humidity (RH), provide early indicators of plant stress, but conventional analytical methods are costly, bulky, and unsuitable for continuous in situ monitoring. Here, a low-cost multimodal sensing platform based on laser-induced graphene (LIG) is reported for real-time, on-leaf detection of methanol, acetic acid, and RH under ambient conditions. The platform integrates Pt-modified LIG electrodes with PtNP/ZnONR@ZIF-8 for methanol, PtNP/ZnONR@ZIF-8/Sn3O4 for acetic acid and GO:PDMAA for RH sensing. After optimization, the sensors respectively achieved sensitivities of −75.83 Ω/log(ppm), −1.63 Ω/ppm, and −4,776.01 Ω/%RH with detection limits of 0.382 ppm, 0.318 ppm, and 1.53 %RH and more than 97% signal retention over 26 days. On-leaf measurements over 2 weeks showed methanol increasing from ~0.3–13.8 ppm to ~13.5–59.0 ppm, acetic acid from ~6.6–14.8 ppm to ~23.3–60.6 ppm, and RH decreasing from ~72.2–86.3% to ~52.5–65.1% under heat stress. These coupled chemical and microclimate changes provide direct, dynamic stress readouts during plant monitoring. By moving beyond single-analyte measurements, the proposed multimodal approach enables early stress detection, data-driven crop management, and next-generation precision agriculture applications.
Potato leaf diseases directly reduce yield and quality, and accurate field detection is important for precision plant protection. However, potato disease lesions are often weak in deep semantic representation, easily disturbed by complex field backgrounds, and variable in multi-scale lesion texture. To address these challenges, this study proposes an improved YOLO-based potato leaf disease detection model. The proposed model enhances the detector through three task-oriented modules. Deep Symptom Enhancement is used to strengthen deep disease feature extraction. Lesion Selection Attention based on large separable kernel attention improves the spatial selection of lesion regions. Multi-Scale Refinement Adapter uses a Mona-based C2PSA structure with two stacked Mona adapters to refine multi-scale texture and lesion-boundary information. Experiments were conducted on a potato leaf disease image dataset using mAP50, average recall (AR), parameters, GFLOPs, and FPS as evaluation metrics. The baseline YOLO26s achieved 81.31% mAP50 and 77.85% AR. The proposed SLR-YOLO model achieved 88.92% mAP50 and 83.51% AR, improving mAP50 and AR by 7.61 and 5.66 percentage points, respectively, while maintaining 118.6 FPS. The results show that the proposed framework improves detection accuracy for potato leaf disease images while retaining practical real-time performance.
Abstract One of the most important challenges faced by smallholder farmers in the agricultural industry is the lack of accurate, timely knowledge to predict and detect crop health issues. Crop productivity is often threatened not only by diseases but also by environmental and physiological stresses, which contribute to significant yield losses and negatively impact the national economy. Traditional detection methods are time-consuming, costly, and require expert knowledge, creating a need for automated and intelligent systems. This work proposes a dual-functional framework that combines crop disease and stress detection using advanced machine learning and deep learning techniques to accurately classify healthy and diseased leaves. This integrated system ensures early detection, reduces crop loss, improves productivity, and provides a scalable, farmer-friendly solution for sustainable agriculture.
Abstract Background Stomata and pavement cells are fundamental components of the leaf epidermis, jointly regulating gas exchange, water loss, and leaf surface expansion. Stomata size, aperture, density, and pavement cell morphology are critical parameters for assessing plant transpiration efficiency, epidermal growth dynamics, and adaptive responses to environmental constraints. Despite their biological importance, quantifying stomatal and pavement-cell traits remains seldom not generalized, simple, and fast enough . Manual or semi-automated approaches limit large-scale phenotyping and restrict the integration of epidermal morphology into crop-improvement pipelines aimed at developing climate-resilient varieties with optimised stomatal patterning. To address such limitations, we developed Stomatalia , a deep learning-based platform designed to automate and standardise the quantification of stomatal and pavement cell traits. The algorithm was trained on epidermal images of cultivated and wild potato and tomato genotypes grown under optimal and abiotic-stress conditions. Stomatalia automatically detects stomata and pavement cells and extracts a broad range of morphological and integrative epidermal parameters, enabling high-throughput phenotyping within a unified workflow. Results Prior to platform development, we optimised a rapid, minimally-destructive leaf-printing protocol that generates negative impressions of the leaf surface within 40–100 s. Transparent positive prints were subsequently produced and imaged under a light microscope at scale settings ranging from 20 to 200 μm. The resulting images are analysed using Stomatalia’s user-friendly web-based interface, which runs an instance-segmentation deep learning algorithm to detect, count, and calculate stomatal and pavement cell parameters. The platform outputs structured files containing raw measurements, derived integrative traits, and associated metadata, facilitating downstream statistical and physiological analyses. Algorithm evaluation on independent datasets demonstrated high performance within the validated dicot imaging domain, with F 1 -scores ranging from 0.86 to 0.94 depending on image scale, species, and resolution, and high segmentation overlap for both stomata and pavement cells. The generality of stomatal detection was also tested on spring onion, chickpea, balsam poplar, and wheat in cross-species feasibility tests, although performance was more variable in monocots, and pavement-cell segmentation remained species- and architecture-dependent. Benchmarking against another publicly available app further showed that, under the tested web interface settings and image types, Stomatalia exhibited closer agreement with manual counts and substantially faster processing times. The practical performance of Stomatalia was further tested in a proof-of-concept trial with potato plants subjected to optimal irrigation and a long, gradual drought. The platform reliably quantified epidermal traits despite variations in leaf morphology and image quality, supporting the integrated interpretation of stomatal and pavement-cell responses under stress. Conclusions We developed Stomatalia as a robust, user-friendly deep learning platform for automated, high-throughput analysis of bright-field leaf epidermal images across varying magnifications and resolutions. Stomatalia facilitates rapid, reproducible, and coordinated phenotyping of stomatal and pavement cells by integrating methodological standardisation, computational automation, and multi-trait extraction in a single analytical workflow. Its strongest current application is the analysis of high-quality dicot leaf-print images, particularly in species and imaging conditions similar to those used for model training and validation. Cross-species and benchmark analyses further define its current scope: stomatal detection can be transferred to some additional epidermal architectures, whereas robust pavement-cell segmentation in monocots or highly divergent species will require further annotation and model retraining. Within these defined boundaries, Stomatalia provides a flexible and extensible framework for studying stomatal and pavement cell morphology and environmental plasticity, while also supporting broader efforts to dissect and optimise plant responses to abiotic stress.
Reproduction assets foundThe authors publicly deposited the paper's test image datasets (raw/input leaf-print images, detection outputs, manual ground-truth counts, exported datasets) and the model file on Figshare, and provide a public Google Colab demo for running the Stomatalia algorithm. Both are paper-specific, public, and actionable.Dataset · publicThe test datasets and model file used in this work are available through the following link:
https://doi.org/10.6084/m9.figshare.32532672. The test_sets.zip archive contains the test image datasets
(cross-species and benchmark analysis), including raw/input images, detection output images, manual
ground-truth counts and exported datasets.Open asset ↗Figshare · 10.6084/m9.figshare.32532672pdf-page:25 lines:1-75Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Grapevines are economically vital crops but are highly susceptible to fungal, bacterial, and viral diseases that threaten yield and quality. Traditional detection methods rely on manual inspection, are time-consuming, prone to human error, and often delay intervention. This study introduces a lightweight convolutional neural network (CNN) architecture specifically designed for accurately and efficiently detecting grapevine leaf diseases—including Black Rot, ESCA, and Leaf Blight—based on image classification. The proposed model integrates optimized residual blocks, batch normalization, dropout layers, and global average pooling to maximize accuracy while minimizing computational complexity. This lightweight design makes it well-suited for deployment on edge devices such as drones and mobile systems used in precision agriculture. A comprehensive data augmentation strategy was applied during training to simulate real-world variability and enhance generalization. The model was trained using 9,027 labeled grape leaf images from a publicly available grape disease image dataset, and it achieved 99.8% overall accuracy with near perfict precision, recall, F1-score, and area under the ROC curve (AUC) across all classes. These findings highlight the practical potential or real-time, scalable and sustainable disease monitoring in smart vineyard management systems.
This study compared unsupervised and supervised machine learning, and deep learning (U-Net) classifiers on Unmanned Aerial Vehicle (UAV) multispectral imagery to identify nitrogen status in potato crops under nitrogen (N) fertilization treatments, irrigation (I), and their interaction (N × I). The U-Net model outperformed all other methods, achieving accuracies for crop nitrogen status of 65–99% in N, 84–100% in I, and 41–82% in N × I treatments, with variation due to different input data. Supervised machine learning also performed well, with Support Vector Machine achieving 53–87, 66–86, and 32–66% respectively, and Random Forest 61–96, 70–81, and 33–65%. Unsupervised K-means yielded the lowest accuracies (47–58, 9–65, and 8–34%), demonstrating necessity of substantial supervision to delineate crop nitrogen and water status. These findings were confirmed by repeated analyses of UAV imagery acquired later in the growing season with consistent results. Comparable classification performance was observed for crop water status and leaf area index at both time points. Despite being demonstrated in a single-field, single-crop framework, the results provide proof of concept for applying deep learning classifiers to detect subtle nitrogen and water stress under field conditions in precision agriculture. Future research could test diverse agroecosystems and growing seasons, alternative deep learning algorithms, and sensor data fusion to improve classification accuracies.
This study aims to improve the performance of the Convolutional Neural Network (CNN) algorithm in detecting coffee leaf diseases using the Inception V3 architecture. The main challenge of this classification is the subtle visual similarity in color, texture, and symptom patterns between diseases. To overcome this, Inception V3 is implemented because of its superiority in multi-scale feature extraction through convolution factorization which reduces parameters while increasing accuracy. The dataset used consists of 1,120 images, evenly distributed into four classes (three types of diseases and one healthy class, each with 280 images), with a training, validation, and test data split ratio of 896:112:112. As a comparison, a conventional basic CNN architecture consisting of 3 convolution layers (3 X 3, stride 1), 3 max-pooling, and 1 dense layer, trained with the same hyperparameters (Adam optimizer, learning rate 0.001, batch size 32) is used. The experimental results show a significant performance improvement; Model accuracy increased from 74.4% on a standard CNN to 97.0% after integrating Inception V3. The scientific contribution of this research lies in mapping overlapping visual characteristics of coffee diseases through multi-scale feature optimization, which demonstrates that computational efficiency can go hand in hand with accuracy improvements on complex agricultural image datasets. These findings confirm that the Inception V3 architecture provides a robust and efficient solution for automating plant disease diagnosis in the field.
Plants can suffer a number of diseases that impact agricultural productivity and food security, particularly in developing farming communities. Although deep learning is capable of classification of diseases with outstanding results, its use is limited due to the difficulty of obtaining large labeled databases and the high requirement of computation. To address these challenges, this study proposes a new method for plant disease classification based on traditional image processing and machine learning algorithms with lightweight and low computation requirements. This one uses several handcrafted descriptors such as color histograms, Haralick texture features and Hu moments to retrieve the information relevant to a disease from the segmented leaf images. Performance of top five classifiers, namely Random Forest, Support Vector Machine, K-Nearest Neighbors, Logistic Regression and Naïve Bayes classifiers are evaluated from the dataset of healthy plant leaves and diseased plant leaves images on 10-fold cross validation. Based on the results of the research work, the best classification model was the Random Forest Classifier model with the accuracy value is 98.12%, 0.98 precision, 0.98 recall, and 0.98 F1 value. The proposed solution was also found to be uncomputation complex and low memory consuming and can be made realtime inference. Therefore, this solution can be implemented onto agricultural systems, considering edge computing and IoT. The results of the research also demonstrated that feature-based machine learning approaches afford interpretable and reliable plant disease detection at a low computation cost, further contributing to sustainable, and precision agriculture.
Mango (Mangifera indica L.) is among the most important commercial fruits grown throughout the world in the tropical and subtropical areas. Even though mangoes are economically important, their cultivation is continuously threatened by a wide variety of leaf and fungi diseases, resulting in crop losses of up to 15-30% annually. [1]. Conventional disease identification depends heavily on expert visual inspection—a process that is inherently slow, subjective, and largely impractical for smallholder farmers operating in remote areas with limited access to agronomic specialists. This paper provides an end-to-end deep learning-based approach towards automatic mango leaf disease detection along with an Android application for real-time deployment in the field. This work is built upon MangoLeafBD [1], an openly accessible dataset that consists of a total of 4,000 images of RGB color space for seven different disease categories and one healthy class, with each category having 500 samples collected from four separate orchards in Bangladesh. Three types of transfer learning models including VGG16 [2], MobileNetV2 [3], and DenseNet121 [4] were considered after applying two-phase fine-tuning based on pre-trained ImageNet weights. For each of the three types of neural network models tested, a series of preprocessing steps consisting of bilinear resizing to size 224 x 224, channel-wise normalization, and image augmentation (rotation, zoom, brightness adjustment, horizontal flip, and shear) was used to increase model accuracy for diverse real-world images. The training model was deployed using TensorFlow Lite (TFLite), which allowed it to be run in offline mode on mid-end Android phones, without needing internet connectivity. The app lets farmers upload leaf images or take images and get a diagnosis of the leaf diseases, along with possible treatments for them.This work contributes a replicable pipeline linking state-of-the-art deep learning research with practical precision agriculture, particularly for rural communities that currently lack access to timely agronomic advisory services.
Accurate detection of plant leaf diseases is essential for enhancing crop productivity and supporting global food security. In addition to disease classification, understanding how environmental and soil conditions affect model performance is important for developing robust real-world agricultural monitoring systems. Although deep learning (DL) models achieve high accuracy on benchmark datasets, their performance in real-world settings is often limited by variations in illumination, background complexity, and environmental conditions. This study proposes a smart DL framework for detecting and classifying multiple leaf diseases in tomato, potato, and pepper plants. The framework combines U2-Net-based leaf segmentation with a Convolutional Neural Network–Bidirectional Gated Recurrent Unit (CNN–Bi-GRU) architecture. MobileNetV2 is employed as the feature extraction backbone to capture spatial characteristics, while Bi-GRU layers model sequential feature dependencies, forming a spatio-temporal network whose architectural design prioritizes parameter efficiency through depthwise separable convolutions and reduced gating complexity. The model was trained and validated using the PlantVillage benchmark dataset and achieved a classification accuracy of 99.8% with a macro-averaged F1-score of 94%, outperforming several state-of-the-art architectures. To assess robustness under real-world conditions, the trained model was further tested on leaf images collected from open-field environments near Eluru, South India. The field evaluation revealed a reduction in classification accuracy to 61.97%, indicating the impact of domain shift and environmental variability. To investigate potential contributing factors, soil parameters, including pH, temperature, moisture, and NPK levels, were monitored using an IoT-based Arduino sensing system over ten consecutive days. Rather than serving as direct inputs to the disease classification model, these environmental measurements were analyzed to assess their potential influence on disease symptom expression and the observed reduction in model performance under field conditions. The results suggest that environmental conditions may influence disease symptom expression and model transferability. This study highlights the importance of integrating DL-based disease recognition with environmental monitoring for reliable field-level agricultural applications. Nevertheless, computational complexity metrics, including inference latency and memory footprint, were not evaluated in the present work and are identified as a priority for future edge deployment studies.
Plant varieties are essential for the survival of human beings and animals, as they act as an alternative source of food, fiber, fodder, and other raw materials for domestic needs and industries in any society. Early identification of plant leaf diseases is very important for keeping the health of crops intact. Crops being infected can affect the overall yield of crops, which may be detrimental to the earnings of farmers. With the emergence of artificial intelligence technology, it has become possible to deploy systems for quicker identification of illnesses. This work has been carried out for the prediction of plant diseases based on visual phenotypic manifestations, such as images of leaves. For this purpose, the dataset has been created after retrieving data from the PlantVillage dataset. The clinical reliability of different deep learning models of various representational capacities has been tested while using ImageNet pre-trained parameters. The experimental results show that MobileNetV2 achieves the highest accuracy of 96.15%, outperforming deep CNN (76.92%) and medium CNN (61.54%). The test accuracy and class-wise F1-scores for the CNN are observed to be substantially very high. The generalization ability and result of DCNN and MCNN are moderate and poor, respectively, as observed. Additionally, the proposed CNN only uses the disease-affected areas on the leaf, thus making the result more interpretable
Cotton leaf diseases present a major threat to global cotton production, significantly impacting both yield and fiber quality. Traditional diagnostic methods are labor-intensive, time-consuming, and demand highly skilled professionals, making them inefficient for large-scale agricultural applications. Although earlier deep learning -based approaches have shown promising results in identifying cotton leaf diseases such as Bacterial Blight, Fusarium Wilt, and Curl Virus Disease, their performance is often limited by complex preprocessing requirements and insufficient generalization to real-world field conditions. To address these challenges, this study proposes and optimized transfer learning-based model, CLDP-CNN, designed to enhance feature extraction and classification efficiency using pre-trained deep neural networks. This study demonstrates the development of Cotton Leaf Disease Prediction Convolutional Neural Network (CLDP-CNN) automatically, utilizing Transfer Learning (TL) which operates on meticulously prepared datasets. Two distinct datasets were used to train the model: the first consisted of field images from cotton farms, while the second was sourced from Kaggle. The main goal of this research examines how the model performs on real-world field datasets. The CLDP-CNN model has proven highly accurate by attaining 99.78% detection success rates for cotton leaf diseases when processing primary dataset which surpasses its secondary dataset accuracy rate of 99.62%. Both the primary dataset and secondary dataset resulted in high accuracy values for the VGG16 pre-trained model which achieved 99.56% accuracy on the primary dataset and 98.82% on the secondary dataset. A web-based application enhances the capabilities of the CLDP-CNN model by providing real-time updates on the health status of cotton plants. This technology empowers farmers with valuable information, enabling them to take timely protective actions to prevent potential severe yield losses in their cotton crops.
Reproduction assets foundThe paper's cotton leaf disease image datasets are publicly available: the authors' primary field-collected dataset on the first author's GitHub repository, and the secondary Kaggle dataset used for comparison. No analysis code or trained model checkpoints are explicitly deposited.Dataset · publicbia.
Funding: This work was supported by Princess Nourah bint Abdulrahman
University Researchers Supporting Project number (PNURSP2026R760),
Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.
Data and code availability
The data that support the findings of this study are openly available in Github and
Kaggle at. https://github.com/mnaeem303/Cotton-Leaf_Disease-Dataset),
and https://www.kaggle.com/datasets/seroshkarim/cotton-leaf-disease-dataset/data
Author Contributions
All the authors (Muhammad Naeem, Muhammad Ibrahim, Nadeem Sarwar, Oumaima
Saidani, Asma Irshad, Muhammad Shadab Alam Hashmi, Muhammad Tayyab
Qammar) contributed equally to this work in their respective meaningOpen asset ↗https://github.com/mnaeem303/Cotton-Leaf_Disease-Datasetpdf-raw-page:29 lines:1-54Dataset · publicbdulrahman
University Researchers Supporting Project number (PNURSP2026R760),
Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.
Data and code availability
The data that support the findings of this study are openly available in Github and
Kaggle at. https://github.com/mnaeem303/Cotton-Leaf_Disease-Dataset),
and https://www.kaggle.com/datasets/seroshkarim/cotton-leaf-disease-dataset/data
Author Contributions
All the authors (Muhammad Naeem, Muhammad Ibrahim, Nadeem Sarwar, Oumaima
Saidani, Asma Irshad, Muhammad Shadab Alam Hashmi, Muhammad Tayyab
Qammar) contributed equally to this work in their respective meaningful ways. All the
authors have read and approved the final manuOpen asset ↗pdf-raw-page:29 lines:1-54Plant phenotyping relevance matchEurope PMC · checked 15 Sept 2026
LeafGrowth / time-series analysisGrowth / development / phenology
A growing body of literature has illustrated the importance of understanding the circadian clock due to its connection to a wide array of molecular and physiological processes. Numerous methods have been developed in order to monitor the status of the circadian clock in living tissue; however, most methods are either costly or labor intensive, and often require precise conditions that lowers throughput and limits flexibility in the size, age, or type of plant to be assayed. Here, we present an affordable and adaptable methodology for assaying circadian period using the well-characterized circadian output of leaf movement. Employing fully automated time-lapse photography using Raspberry Pi cameras and predominantly automated image post-processing, this methodology minimizes manual input to expedite circadian analysis, thus improving throughput. Additionally, the top-down setup used in this method is appropriate for a wide range of sizes and ages of plants, allowing for an expansion of the scientific questions that can be assayed by this methodology.
Early-generation sesame (Sesamum indicum L.) breeding requires high-throughput phenotyping of large unreplicated populations across contrasting environments. A DJI Phantom 4 Multispectral UAV was flown at 40, 80, and 120 m above ground level (AGL) over 588 M2 genotypes under full irrigation (ENV1) and terminal drought (ENV2; irrigation withheld from reproductive onset) on four dates (July–September 2025). Structure-from-motion canopy height models were compared with ground measurements, and four spectral reflectance indices—Normalised Difference Vegetation Index (NDVI), Normalised Difference Red Edge (NDRE), Green Normalised Difference Vegetation Index (GNDVI), and Leaf Chlorophyll Index (LCI)—were derived from 40 m imagery. Ordinary least squares (OLS), Random Forest, and Gradient Boosting were evaluated under leave-one-genotype-out (LOGO), leave-one-environment-out (LOEO), and leave-one-date-out (LODO) cross-validation; genotypic repeatability was quantified by intraclass correlation (ICC), and drought performance was ranked by a composite Stress Tolerance Index (STI) validated against an independent breeder assessment. The 40 m altitude gave the highest height accuracy (R2 = 0.812 in ENV1; 0.663 in ENV2). LOGO accuracy (R2 ≈ 0.83) fell to R2 ≈ 0.55 under LODO—the operationally relevant figure for a new phenological stage—and the full structural–spectral OLS model collapsed (R2 = −0.203) where tree ensembles remained stable. Spectral-index repeatability was up to ~2-fold higher under stress (ICC(3,4) > 0.84). The composite STI flagged 38 elite genotypes (7.6% of 498); 10 of its top 30 were confirmed in the breeder’s 48-best selection from all 588 rows—a 4.1-fold enrichment over chance (hypergeometric p = 4.5 × 10−5).
NeRF / 3D Gaussian SplattingLiDAR / point cloudLeafWhole plant / canopy / plot / field2D/3D reconstructionGrowth / time-series analysisTrackingArchitecture / morphology / geometryGrowth / development / phenology
Quantifying plant growth dynamics from sparse longitudinal 3D observations is fundamental for agriculture and plant sciences. Yet, plants pose unique challenges: they undergo intricate non-rigid deformations, exhibit changing topology as new organs emerge, and often lack explicit temporal correspondences between consecutive data acquisitions due to newly formed tissue. Methods designed for general scenes struggle to model topology changes and asynchronous organ growth characteristic of plants. To address these challenges, we introduce GrowFields, a compositional dynamic neural field representation for organ-aware 4D plant growth modelling from point cloud time series. Our approach decomposes a plant into its constituent organs and aligns each organ into its own canonical coordinate frame, isolating intrinsic growth patterns from global plant motion. We then learn a shared continuous neural deformation field that models temporal dynamics across all organs, conditioned on learnable per-organ latent codes capturing organ identity and growth characteristics. The resulting modular yet unified representation naturally accommodates the asynchronous development of plant organs while remaining grounded in the practical setting of organ-level plant tracking. We evaluate GrowFields on growth sequences from four plant species, assessing geometric fitting and organ tracking accuracy using manually annotated leaf-tip trajectories. Results demonstrate consistent improvements in spatial precision, temporal coherence, and morphological fidelity over a range of existing representations.
Food productivity, quantity and quality are at stake when plant diseases such as rice diseases undermine the food security. Rice leaf disease treatment necessitates accurate and timely diagnosing. This study describes a deep learning model for categorizing and forecasting rice plant diseases. Using the remora optimization algorithm (ROA) on a rice leaf dataset demonstrates its potential for plant disease classification. The ROA-DM method detects rice leaf diseases using the ROA algorithm, a deep maxout network (DMN), and a deep autoencoder (DAE). ROA is applied to the learning parameters of deep model in order to achieve better convergence and avoiding local minima, which usually happens with conventional gradient-based optimizers. Experiments show that the suggested framework is accurate and precise across illness categories. The confusion matrices display the training and validation accuracy, losses of this model. The performance of our optimal learning method with respect to other methods indicated its potential for identifying leaf diseases. The accuracy of the ROA-DM method is 98.5%.
Reproduction assets found保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。Dataset · publicThe dataset collected from https://www.kaggle.com/datasets/emmarex/plantdisease (PlantVillage dataset) for algorithm testing in plant disease diagnosis 35 . The selected rice leaf disease samples from PlantVillage dataset consisting of 3050 colour leaf images across four classes.Open asset ↗Kaggle · emmarex/plantdiseaselines:85-97Plant phenotyping relevance matchCrossref · Europe PMC · checked 15 Sept 2026
Introduction Accurate and rapid diagnosis of plant leaf disease symptoms is critical for sustainable agricultural crop production, yet traditional methods often lack efficiency and robustness under field conditions. Methods Here, we propose a deep learning framework based on an improved Vision Transformer architecture that integrates a dynamic sparse attention mechanism, termed KBTNet, for targeted feature extraction in symptom-affected regions of leaf images. The model incorporates a learnable nonlinear enhancement module to capture subtle visual disease symptom variations such as lesions, discoloration patterns, and spot distributions, and a lightweight Transformer design to reduce computational cost. Results Evaluated on a multisource dataset containing soybean and tomato leaf images representing diverse disease symptom patterns, our approach achieved 93.19% classification accuracy, outperforming current state-of-the-art models. Additional evaluations on public plant disease datasets from multiple crops further demonstrate the model's ability to recognize disease symptom patterns across diverse crop species. Discussion The proposed framework achieves stable performance across diverse crop and disease symptom categories, maintains high efficiency under reduced parameter complexity, and exhibits strong potential for realtime field diagnostics on edge devices. This work provides a scalable and efficient tool for plant disease symptom detection and classification and supports the integration of visionbased intelligence into crop disease monitoring and management systems.
Rice is the staple food for more than half of the global population. However, the productivity and quality of rice or grain dropped significantly due to leaf diseases. These diseases are difficult to identify through manual processes, which are time-consuming, labour-intensive and often inaccurate, particularly in rural farming communities. With recent advances in computer vision technology, object detection algorithms, namely the you only look once (YOLO) family, can provide high-speed, high-accuracy solutions for automated plant disease detection. This study evaluates four YOLO variants, such as YOLOv5, YOLOv7, YOLOv8 and YOLOv11, using about 1500 field images collected in Bangladesh. The data represent four major rice leaf diseases, such as bacterial leaf blight, brown spot, leaf blast and sheath blight. Data pre-processing, including image annotation and data augmentation, was conducted before model training and was followed by the training of the YOLO models. All the models were trained with the same hyperparameters and their performance was evaluated using standard metrics, such as F1 scores, precision, recall and mean average precision (mAP). According to experimental findings, the YOLOv7 recorded the highest performance based on F1 score of 0.77 and mAP of 0.85 in comparison with the rest of the variants. The results suggest that YOLOv7 will be the most appropriate to use instead of other models in the detection of rice leaf disease in real-time, which can be utilised in precision agriculture and mobile-based disease management systems.
Plant diseases remain a major challenge to global food production, and timely, accurate, and scalable detection of plant stress is critical to reducing these losses. Recent advances in digital imaging and artificial intelligence offer unprecedented opportunities for precision crop disease detection and management. Yet, existing plant disease datasets remain often fragmented across crop and disease systems, and are largely dominated by controlled-environment imagery. The lack of standardized, interoperable, and representative datasets limits reproducibility, transferability, and scalability of AI systems, thereby constraining their deployment in operational agricultural applications. Here we present LeafMD, an integrated multimodal plant disease dataset and benchmark resource that includes LeafNet 2.0, a large-scale multimodal digital image dataset comprising 255,855 image–text pairs across 37 crop species, 197 crop–disease classes, and 9 geographic regions spanning tropical, subtropical, and temperate agricultural systems. Unlike conventional datasets, LeafNet 2.0 integrates biologically grounded symptom descriptions with image-level annotations of early and late disease stages, enabling symptom-aware analysis of disease progression under realistic field conditions. We further introduce LeafBench 2.0 as part of LeafMD, a visual-question answering benchmark covering nine fine-grained plant pathology tasks, including pathogen classification, lesion characterization, symptom interpretation, and disease severity assessment. Evaluation across 16 vision–language models revealed substantial performance gaps between coarse disease recognition and fine-grained pathological reasoning, while agriculture-adapted models consistently outperformed several larger general-domain architectures on symptom-oriented tasks. Together, LeafNet 2.0 and LeafBench 2.0 establish LeafMD as a multimodal resource for developing disease-aware agricultural foundation models and studying fine-grained pathological reasoning in real-world environments.
Rice leaf diseases can cause a significant decrease in productivity if not treated early, while currently still using a manual diagnosis process that is often slow, inconsistent, and dependent on extension workers. In this study, the SRIKANDI application was developed, a mobile application for image processing for rice leaf diseases using the NASNetMobile Convolutional Neural Network (CNN) architecture. This system is designed using five labels, namely bacterial, blast, brownspot, leafsmut, and healthy leaves. The dataset used consists of 2500 images collected from Kaggle, Mendeley Data, and taken directly. All images go through preprocessing stages of resizing, pixel normalization, and augmentation, then divided into 80% train, 10% test, and 10% validation. The model training was carried out in two stages, namely, 40 epochs of fine-tuning with a learning rate of 0.0008 followed by 20 epochs of fine-tuning with a learning rate of 1e-5, the results obtained by the model with a test set accuracy rate of 96.40%. The trained model is then saved in TFLite format to be integrated into the SRIKANDI mobile application so that it can help farmers detect rice leaf diseases in real-time via camera or taken from the gallery.
A photoelectrochemical (PEC) aptasensor based on bismuth oxyiodide (BiOI) nanoflower/biomass carbon (BiOI@BC) was fabricated for in-situ detecting abscisic acid (ABA) in tomato leaves under salt stress. Shrimp shells-derived biomass carbon acted as an enhanced carrier, and the biomass carbon improves the PEC performance of BiOI by extending the visible light absorption range and promoting the charge transfer of pure BiOI nanoflower. The BiOI@BC exhibited high photocurrent, which was about 19 times in contrast to pristine BiOI, attributing to the synergistic effects of biomass carbon self-doped with N, P, and S atoms. Furthermore, a PEC aptasensing platform was developed for the sensitive and selective determination of ABA, with a wide linear range from 0.1 to 1000 pM and a remarkably low detection limit of 0.03 pM. The practical applicability of the device was further validated by on-site monitoring of ABA levels in tomato leaves under salt stress, demonstrating good stability and accuracy. This work provides a robust strategy for real-time phytohormone detection, facilitating precise crop regulation in plant biology and agriculture.
Leaf orientation (posture) influences photosynthesis and plant responses to environmental cues. However, the existing methods for quantifying posture typically compress its inherently three-dimensional structure into scalar values or single vectors, leaving the 3D aspects of leaf movement poorly understood. For a more complete geometric description, we propose representing leaf-blade posture by an orthonormal basis (ONB), defined as three perpendicular unit vectors aligned with the developmental axes of the leaf. This ONB serves as a local coordinate system and corresponds to rotation matrices used to represent orientation in three-dimensional space, embedding leaf posture within the mathematical structure of the special orthogonal group SO(3). Using three-dimensional point-cloud data, we reconstructed an ONB aligned with the three axes of the leaf blade and quantified elevation and azimuth angles. When applied to diurnal posture changes, the resulting angular patterns were consistent with previous observations. We then visualized posture changes after gravitational perturbation as continuous rotational trajectories. These trajectories could be compared with mathematically defined geodesic shortest paths and used to simulate alternative reorientation routes that satisfy constraints. The rotational trajectories could also be separated into swing and twist components, which reflect distinct deformation modes. The ONB can be obtained not only from three-dimensional point clouds but also from other measurement tools, making the approach broadly applicable. Conceptually, ONB representation places leaf posture within the geometric structure of SO(3), enabling the use of well-established mathematical tools such as rotational distance and geodesics for analyzing leaf reorientation.
Accurate monitoring of nitrogen nutrition is critical for optimizing cotton production. Traditional machine learning-based inversion models have limited effectiveness for precision monitoring. Multisource fusion models for small samples were developed in this study to achieve enhanced accuracy through fitting and data complementarity. Cotton plants subjected to different nitrogen treatments were investigated. A two-year pot experiment was conducted to collect main-stem leaf images to construct an image pretraining dataset for model transfer. In a field experiment conducted over one year, main-stem leaf data were collected using hyperspectral, chlorophyll fluorescence, and digital camera sources, thereby providing a multisource dataset for training monitoring models. Two architectures—a neural network (NN) and an interpretable deep forest (DF), which are suitable for small-sample spectral, fluorescence, image color, and texture-sequence features—were constructed to improve the accuracy of nitrogen content inversion. Additionally, a two-dimensional sliding-window processing method was introduced into the DF multigranularity scanning module, and a transfer-learning-based two-dimensional convolutional NN was employed to directly model small-sample two-dimensional images. Building upon the outcome, multilayer fusion models were constructed, with corresponding fusion strategies designed for homogeneous sequence inputs and heterogeneous image–sequence inputs. The results showed that NN and DF can effectively handle limited sample sizes and outperform traditional machine learning models. Among the fusion models, the optimal secondary decision-level fusion model achieved an R² of 0.926 on the independent test set, indicating good performance under small-sample conditions. This study provides a methodological reference for the precise monitoring of crop phenotypic parameters under small-sample conditions.
The objective of this study was to develop a 3D plant modeling strategy that enables camera pose recovery from segmented plant images and the reconstruction of an initial point cloud. A lightweight, contour-aware framework leverages the view-consistent and surface-oriented representation of 2D Gaussian Splatting, making it suitable for plant surface reconstruction under the Plant-to-Camera mode. A contour-weighted Laplacian regularization suppresses depth discontinuities around plant boundaries, while simplified Gaussian primitives improve computational efficiency without compromising geometric fidelity. Organ-level semantics are integrated into the reconstructed geometry to distinguish plant organs such as leaves, stems, and ears. On maize and wheat datasets, our method outperformed existing approaches in terms of morphological fidelity, organ-level structural consistency, and processing speed, demonstrating its suitability for plant phenotyping
A. Conte · A. Carli · M. Haworth · G. Marino · V. Montesano · D. Danzi · G. Atzori · A.P.M. Fabbri · A. Daccache · R.M. Balestrini · M. Centritto
Field / plotLeafWhole plant / canopy / plot / fieldPhysiological trait estimationPhotosynthesis / fluorescenceWater status / transpiration
• Rapid methods show reduced robustness in hot summer Mediterranean field conditions. • Method‑dependent differences in Vc max estimates strongly affect A - g s model outputs. • Steady‑state A / C i provides the most accurate gs simulations. • RACiR provides Vc max estimates closest to A / C i and represents a suitable option for high‑throughput phenotyping. The maximum rate of carboxylation of ribulose-1,5-bisphosphate ( Vc max ) represents a key biochemical trait and a fundamental parameter in C3 models of photosynthesis, as it enables an accurate representation of leaf carbon assimilation and gas exchange. Accurate estimation of this parameter is essential for process‑based modelling across scales, as uncertainties in Vc max may influence model behaviour when scaled from leaves to larger spatial domains. Traditionally, Vc max is derived from the response of photosynthesis ( A ) to intercellular CO 2 concentration ( C i ), known as the A / C i curve, a reliable but time-consuming and labour-intensive procedure that limits its application in high-throughput phenotyping. To address this limitation, rapid approaches such as the Rapid A/Ci Response (RACiR) and the one-point (OP) methods have been developed. However, their accuracy, reliability, and reproducibility must be carefully validated, as discrepancies arising from the use of heterogeneous data sources for model parameterization may introduce significant uncertainty. In this study, the RACiR and the OP methods were evaluated against the conventional A / C i curve in a two-year field experiment on four Cannabis sativa varieties grown under different irrigation regimes. Photosynthetic traits derived from each method were compared and integrated into a coupled A -stomatal conductance ( g s ) model to assess how method-driven differences affect model outputs. Overall, photosynthetic traits estimated from A / C i curves provided the most accurate simulations of g s , with R 2 values ranging from 0.55 to 0.84 and the lowest RMSE. In contrast, traits derived from RACiR and OP methods resulted in g s overestimations of 26.7% and 50.7%, respectively. Field application of RACiR was hindered by high failure rates under high summer temperatures, while OP estimates showed substantial variability. These results indicate that, despite the appeal of faster alternatives, the A / C i curve remains the most reliable method for estimating Vc max under Mediterranean field conditions, particularly when high accuracy is required for model-based applications.
To address the insufficient characterization of vertical heterogeneity in potato canopy leaf nitrogen content (LNC), this study developed a layer-specific LNC estimation framework based on canopy hyperspectral reflectance, fractional-order derivative (FOD) transformation, and two-band and three-band optimized spectral indices. Partial least squares regression (PLSR) was then used to evaluate the predictive ability of the selected spectral indices for Top, Middle, and Bottom LNC. Field experiments were conducted from 2022 to 2023 in the semi-arid region of Yulin, Shaanxi Province, China. Canopy hyperspectral reflectance from 350 to 1830 nm and LNC measurements of upper (Top), middle (Middle), and lower (Bottom) leaves were synchronously acquired during the tuber formation stage. The results showed that potato canopy LNC exhibited a clear vertical gradient, following the order Top LNC > Middle LNC > Bottom LNC. Traditional vegetation indices were significantly correlated with LNC, but their correlations decreased with increasing canopy depth, with the highest correlation for Bottom LNC being only 0.524. Compared with traditional vegetation indices, FOD-based two-band indices showed stronger Pearson correlations with layer-specific LNC. Under FOD1.5, the maximum absolute Pearson correlation coefficients (|r|) between the selected two-band indices and LNC reached 0.855, 0.849, and 0.814 for Top, Middle, and Bottom LNC, respectively. The three-band optimized spectral indices further enhanced spectral information extraction, with maximum |r| values of 0.893, 0.885, and 0.852, respectively. However, cross-year validation produced substantially lower R 2 values, indicating limited temporal transferability of the selected indices and the need for further validation before broader application. Compared with the traditional vegetation index model, it increased the testing-set R 2 for Bottom LNC by 0.279 and reduced RMSE from 0.159 to 0.113. These results suggest that FOD1.5-integrated three-band optimized spectral indices can improve the indirect estimation of layer-specific LNC from canopy reflectance, particularly for Bottom LNC, where the reflectance-LNC association is affected by canopy signal attenuation and mixing. The findings provide a methodological reference for describing canopy vertical nitrogen status and functional heterogeneity in potato, while their broader applicability requires further validation across growth stages, cultivars, sites, and nitrogen management conditions.
Genetically encoded biosensors are one of the essential tools in biological research. They enable visualization of molecules of interest from the subcellular level to entire organism level in vivo and can be used to monitor the presence of small molecules, gene expression, protein activity, and protein degradation. However, multiplexing fluorescent biosensors in plants is notoriously difficult due to signal bleed-through and strong autofluorescence from chlorophyll. In this study, we investigated the potential of multiplexing biosensors based on the selection of reporter fluorescent proteins. We characterized the emission spectra, fluorescence lifetimes, and relative brightness of diverse fluorescent proteins in plant leaves. We show that selected proteins exhibit comparable brightness, supporting their use in co-expression experiments and reliable quantification of individual signals. To separate three overlapping signals, we applied two different linear unmixing approaches and compared them to results obtained without unmixing. We identified the channel separation unmixing approach as the most suitable for biosensors. Additionally, we show how unmixing with the selected approach can be applied to separate autofluorescence and five fluorescent proteins. We further validated this approach in virus-infected cells by following organelle dynamics in vivo. Finally, we demonstrate the feasibility of high-throughput segmentation and quantification with a custom MATLAB workflow for nuclei, chloroplasts, and cytoplasm signal analysis. Overall, our work demonstrates that biosensors can be multiplexed, even when their emission spectra overlap.
Reproduction assets found保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。Code · publicThe segmentation and histograms were acquired using MATLAB script ( https://github.com/NIB‐SI/Nuclei‐segmentation ). The parameters used in the script to achieve appropriate segmentation are listed on GitHub, Case 1 ( https://github.com/NIB‐SI/Nuclei‐segmentation ).Open asset ↗NIB‐SI/Nuclei‐segmentationlines:255-341Plant phenotyping relevance matchCrossref · checked 6 Sept 2026
Chlorophyll content represents a key growth indicator for maize. The traditional SPAD (Soil and Plant Analyzer Development) method, though easy to operate, is inefficient, destructive, and unsuitable for high-throughput field monitoring. UAV (Unmanned Aerial Vehicle) remote sensing technology is highly efficient and detects abundant indicators, enabling large-scale SPAD measurement. In this study, 18 vegetation indices and eight texture features were selected as the indicator system by combining prior knowledge and experimental analysis. In a two-year maize density experiment, multispectral images were collected in the growth period. The correlations among SPAD values, multispectral indices and texture features were analyzed using Pearson correlation coefficients. Then the detection accuracies of three algorithms, i.e., RF (Random Forest), PLSR (Partial Least Squares Regression), and SVR (Support Vector Regression), were compared under this indicator system. Compared with models constructed using single vegetation indices or single texture features, the estimation accuracy of the indicator system at the jointing stage was improved by 0.13 and 0.22, respectively. The results showed that SVR achieved the highest estimation accuracy among the three algorithms, with determination coefficients (R2) of 0.73, 0.77and 0.70 at the jointing, silking, and grain-filling stages, respectively. This study established a non-destructive monitoring framework for chlorophyll content during the entire maize growth stage based on UAV data.
Crop nutrition deficiency poses a major challenge to achieving optimal yield, particularly in smallholder farming systems where timely expert diagnosis is limited. Early detection is crucial to minimize losses and reduce unnecessary fertilizer or pesticide usage. While deep learning offers potential for automated visual diagnosis, most existing approaches operate as black boxes and lack interpretability, explainability, or actionable recommendations. In this work, we present a neurosymbolic framework for early nutrient deficiency detection in ash gourd leaves using the EarlyNSD dataset. Our approach integrates a ResNet-50 backbone with a dual-head design: a classification head for deficiency prediction and a concept-prediction head that quantifies physiologically meaningful visual patterns such as yellowing, edge discoloration, spots, and vein greenness. These concept scores are combined with predefined domain rules to guide the learning of the neural component and to generate transparent, human-aligned explanations for each diagnosis. Building on the model outputs, we incorporate a Retrieval Augmented Generation (RAG)-based pipeline along with an agricultural knowledge base to generate targeted recommendations. This approach overcomes key shortcomings of pure neural models by incorporating domain knowledge in the form of differentiable fuzzy logic rules. The study demonstrates that the proposed framework improves both classification performance and interpretability compared to standard ResNet baselines. Grad-CAM analysis demonstrates that concept-guided attention aligns with symptom-specific regions, such as yellowed areas for Nitrogen deficiency or marginal discoloration for Potassium deficiency, providing visual validation of the reasoning process. Since EarlyNSD is limited in scale and visual diversity, the results are not directly comparable to large open-field datasets. Overall, our results establish a proof of concept for integrating neural detection with symbolic reasoning, enabling interpretable, actionable, and domain-informed nutrient management for practical applications.
Zhao-Kui Li · Wen Cheng · Xue-Wei Gong · Qing-Song Yu · Heng-Fang Wang · Zhong-Yi Pang · Yan-Hui Peng · Xue-Kai Sun · Ming-Yong Li · Guang-You Hao
Field / plotLaboratory / benchtopLeafPhysiological trait estimationSegmentationWater status / transpiration
Abstract Climate-change-driven drought intensification increasingly threatens forest ecosystems, highlighting an urgent need for accurate monitoring of forest water stress. Leaf water potential (Ψleaf) is a key integrative indicator, yet conventional measurements are destructive and unsuitable for large-scale or high-frequency monitoring. Hyperspectral remote sensing offers a promising alternative, but robust canopy-level Ψleaf estimation remains constrained by limited labeled data and heterogeneous environmental conditions. Here, we develop a cross-scale framework integrating supervised contrastive learning with deep transfer learning to translate robust leaf-scale pretraining into canopy-scale Ψleaf estimation from hyperspectral data in a Populus × euramericana ‘I-214’ plantation. Hyperspectral imagery was captured at the leaf scale under controlled laboratory conditions (n = 229) and at the canopy scale using a UAV-based platform (n = 200), together with paired Ψleaf measurements. Reflectance consistently increased with declining Ψleaf at both scales, supporting the feasibility of cross-scale modeling. At the leaf scale, physics-consistent spectral augmentation coupled with contrastive learning enhanced feature discrimination and predictive stability under small-sample conditions (R2 = 0.8030). Transfer learning via progressive fine-tuning enabled efficient scaling of the leaf-level pretrained model to canopy-level prediction despite structural and environmental complexity and restricted field data ranges, achieving R2 = 0.7605 and RMSE = 0.1056 MPa. Coupling with individual-tree crown segmentation further enabled spatially explicit mapping of canopy Ψleaf and plot-level forest water stress dynamics. These results demonstrate that combining contrastive representation learning with cross-scale transfer provides a practical pathway for physiological monitoring and scalable, climate-smart forest phenotyping in data-constrained forested environments.
Rice leaf diseases pose a major challenge to crop health and agricultural productivity, particularly when timely and accurate diagnosis is required under natural field conditions. The development of automated disease recognition systems depends heavily on the availability of large, well-annotated image datasets. However, many existing rice leaf disease datasets are limited in terms of environmental variability, disease representation, and real-field imaging conditions. To address this gap, this paper presents BanglaRiceLeaf, an original rice leaf image dataset collected and curated by the authors from the experimental fields of the Bangladesh Rice Research Institute (BRRI), Gazipur, Bangladesh, between July 2023 and July 2024. The dataset contains 4152 images belonging to five classes: Bacterial Leaf Blight, Bacterial Leaf Streak, Sheath Blight, Leaf Blast, and Healthy Leaf. The images were acquired from two rice varieties, BR11 and BRRI dhan34, under natural field conditions across varying illumination environments in order to reflect practical disease recognition scenarios. All images were manually annotated by trained annotators under expert supervision. The dataset is systematically organized and publicly released to support reproducible research in rice disease classification. In addition to dataset presentation, benchmark experiments using Xception, NASNetMobile, and InceptionV3 are provided to demonstrate its applicability for deep learning-based disease recognition. BanglaRiceLeaf is expected to serve as a useful resource for plant disease analysis, comparative model evaluation, and future research in precision agriculture and agricultural computer vision.
Reproduction assets foundThe paper's core asset is the BanglaRiceLeaf rice leaf disease image dataset (4152 field images, five classes), publicly released on Harvard Dataverse with DOI 10.7910/DVN/XAOBYW. No author analysis code or trained model checkpoints are stated as publicly available.Dataset · publicData Identification Number: https://doi.org/10.7910/DVN/XAOBYW
Direct URL to Data: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XAOBYW
Access Instructions: This dataset is publicly available on the Harvard Dataverse repository and can be accessed for academic, research, and instructional purposes.Open asset ↗Harvard Dataverse · doi:10.7910/DVN/XAOBYWhtml-lines:100-131Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Abstract Early and accurate detection of plant diseases is vital for global food security and sustainable agriculture. While deep learning offers promising solutions, there is a continuous need for architectures that enhance learning capacity and efficiency. This study introduces ViT-KAN, an innovative hybrid model merging the powerful feature extraction of Vision Transformers (ViT) with the flexible, learnable activation functions of Kolmogorov-Arnold Networks (KAN). By replacing the standard Multilayer Perceptron (MLP) classification head of ViT with a KAN module, the proposed architecture aims to better capture nonlinear patterns in agricultural images. Evaluated on the PlantVillage dataset for potato and maize leaf diseases using standard fivefold cross-validation, with final results reported as mean ± standard deviation across the five folds, the model was trained entirely from scratch. ViT-KAN achieved 99.49 ± 0.13% accuracy on the maize dataset and 98.28 ± 0.51% on the potato dataset, compared with 98.92 ± 0.40% and 97.77 ± 0.88%, respectively, for the standard ViT model. Beyond mean accuracy, ViT-KAN showed lower standard deviation across folds, while representative fold curves suggested smoother early training trajectories under the shared training configuration. These findings suggest that ViT-KAN is a promising alternative to conventional ViT-based classification models for plant disease diagnosis.
Crop disease identification is still a big problem in agriculture, which results in large yield losses and food lacks, especially in regions dependent on manual monitoring. Traditional methods of identifying plant diseases are often labor intensive, error-prone, and ineffective in early-stage diagnosis. To overcome these limitations, this study proposes a hybrid machine learning model for accurate crop type identification, disease classification, and severity prediction using image data. The methodology utilizes the PlantVillage dataset, encompassing over 50,000 annotated leaf images across 14 crops. After rigorous preprocessing involving image resizing, normalization, and cleaning, Improved Deep Joint Segmentation is applied to localize disease-affected regions. Feature extraction incorporates color, texture (GLCM, LBP), and shape attributes to enhance classification accuracy. A hybrid approach integrating XGBoost for feature selection and Support Vector Machine (SVM) for classification is proposed to capture both overarching trends and intricate details. Experimental results across four major crops—potato, tomato, corn, and grape—demonstrate superior performance, with the hybrid model achieving 98.6% accuracy, 98.3% precision, 99.0% recall, and 99.1% F1-score. The model outperforms existing approaches, offering a robust, scalable, and accurate solution for early crop disease detection in precision agriculture.
Agriculture is one of the most important sectors contributing to economic development and global food security. However, plant diseases caused by fungi, bacteria, viruses, and other pathogens significantly reduce crop yield and quality, leading to substantial economic losses for farmers. Early and accurate identification of plant diseases is essential for effective crop management and timely application of suitable pesticides. Conventional methods of disease diagnosis rely on manual inspection by agricultural experts, which is time-consuming, labor-intensive, expensive, and often inaccessible to farmers in remote areas. Recent advancements in Artificial Intelligence (AI) and Deep Learning have provided efficient solutions for automating plant disease detection through image analysis. This project, "Plant Disease Identification and Pesticides Recommendation System Using Convolutional Neural Network (CNN) for Crop Protection," presents an intelligent system that automatically identifies plant diseases from leaf images and recommends appropriate pesticides for effective crop protection. The proposed system utilizes a Convolutional Neural Network (CNN), a deep learning model specifically designed for image classification tasks. The CNN model is trained using a large dataset of healthy and diseased plant leaf images collected from publicly available agricultural datasets. During training, the model learns to recognize disease-specific visual features such as color variations, lesion patterns, texture changes, and leaf deformities. Image preprocessing techniques, including resizing, normalization, and data augmentation, are employed to improve the quality of the input images and enhance the overall performance of the model. When a farmer uploads an image of a plant leaf through the system, the trained CNN model analyzes the image and accurately classifies it as either healthy or affected by a specific disease. After identifying the disease, the system recommends suitable pesticides, fungicides, insecticides, or biological treatments based on an agricultural knowledge database. It also provides additional information such as recommended dosage, application method, spraying schedule, safety precautions, and preventive measures to ensure responsible pesticide usage and minimize environmental impact. The proposed system offers several advantages, including rapid disease detection, high classification accuracy, reduced dependence on agricultural experts, optimized pesticide application, lower crop losses, improved productivity, and support for sustainable farming practices. Furthermore, the system can be deployed as a web or mobile application, enabling farmers to access disease diagnosis and treatment recommendations anytime and anywhere using smartphones or other digital devices. Overall, the proposed CNN-based plant disease identification and pesticide recommendation system provides a reliable, cost-effective, and intelligent solution for modern agriculture. By combining image processing, deep learning, and agricultural expertise, the system supports precision farming, enhances decision-making, reduces unnecessary pesticide usage, and contributes to increased crop productivity, environmental sustainability, and long-term food security.
Published1 Jul 2026Microscopy and microanalysis : the official journal of Microscopy Society of America, Microbeam Analysis Society, Microscopical Society of CanadaCited by 0 · OpenAlex ↗
Fuzhou represents a critical center for tea genetic diversity, yet the micromorphological basis for differentiating its local landraces remains poorly understood. Scanning electron microscopy (SEM) was employed to investigate the foliar micromorphology of 28 tea landraces from Fuzhou and to characterize structural differences among them. Adaxial epidermal wax ornamentation, stomatal architecture, and nonglandular trichome patterns provided important taxonomic characters for germplasm classification. Our analysis reveals that stomata are consistently paracytic and randomly oriented on the abaxial surface. However, their dimensions exhibit high phenotypic plasticity, with mean areas ranging from 421.92 to 822.26 µm2. Leaf surface ornamentation showed high phenotypic variability, with three identifiable types: straight, wrinkled, and undulated. The length, width, and type of nonglandular trichomes varied among the landraces, with values of nonglandular trichome length ranging from 269.99 to 632.31 µm and diameter from 9.72 to 14.62 μm. The nonglandular trichome ornamentation was categorized as smooth, long-stripe, and short-stick. The study demonstrated that SEM-based analysis of foliar micromorphological traits provides a valuable tool for tea germplasm identification and cultivar improvement. Specifically, the combination of adaxial epidermal wax ornamentation and nonglandular trichome surface ornamentation provides stable and reliable diagnostic micromorphological markers for accurate differentiation and identification of Fuzhou tea landraces, filling a critical micromorphological gap in the systematic study of local tea germplasm.
Modern agriculture is rapidly adopting Artificial Intelligence (AI) and Internet of Things (IoT) technologies to improve crop monitoring and decision-making. Many existing systems focus either on water stress detection or pest detection separately. The proposed system integrates both functions into a single platform. It uses a camera module and environmental sensors connected to a Raspberry Pi (5/4) as the main controller. A Convolutional Neural Network (CNN) model processes leaf images captured by the AI camera, while a soil moisture sensor supports water stress analysis. The system classifies crops into three categories: healthy, water-stressed, and pest-infected. Based on the output, it provides real-time recommendations for irrigation and pesticide application. This reduces manual inspection, prevents unnecessary chemical usage, saves water, and improves crop productivity.
Reproduction assets foundThe paper's CNN phenotyping/classification analysis is built directly on two public Kaggle image datasets (PlantVillage plant disease and Crop Water Stress), explicitly cited with URLs. No author code or trained model is deposited.Dataset · publicThe PlantVillage Dataset was used for
plant disease detection, and it is available at https://www.kaggle.com/datasets/emmarex/plantdisease.Open asset ↗Kaggle · emmarex/plantdiseasepdf-page:7 lines:1-57Dataset · publicThe Crop Water Stress Dataset was used for crop water stress analysis, and it can be accessed at
https://www.kaggle.com/datasets/harshilsharma/crop-water-stress.Open asset ↗Kaggle · harshilsharma/crop-water-stresspdf-page:7 lines:1-57Plant phenotyping relevance matchCrossref · checked 15 Sept 2026
Published1 Jul 20262026 9th International Conference on Computing Methodologies and Communication (ICCMC)Cited by 0 · OpenAlex ↗
Early and accurate plant disease diagnosis is essential for sustainable crop production and effective disease management. This study presents an Explainable Gradient-Based Convolutional Neural Network (EG-CNN) that integrates leaf image data with gene expression and metabolomics information to enhance disease classification while providing transparent and interpretable predictions. The proposed framework was evaluated on four major plant diseases, powdery mildew, blight, anthracnose, and leaf spot, using a multimodal dataset comprising 10,000 images and associated omics data. Comparative analysis against Traditional CNN, ResNet-50, Vision Transformer (ViT), and CNN-LSTM models demonstrated the superior performance of EG-CNN. The model achieved 97.4% accuracy, 97.1% precision, 96.8% recall, and a 96.9% F1-score, outperforming all benchmark approaches. Paired t-test results revealed statistically significant improvements (p 0.05) over competing models. Confusion matrix analysis indicated minimal misclassification across disease classes, while ROC analysis produced near-perfect AUC values, confirming excellent class separability, sensitivity, and specificity. Five-fold cross-validation further demonstrated robust generalization, with accuracy ranging from 97.1% to 97.6%. Grad-CAM visualizations successfully identified disease-relevant regions, including lesions, discoloration, and necrotic tissues, thereby enhancing model transparency and user trust. The integration of omics data also improved biological interpretability by linking predictions to underlying molecular responses. In conclusion, the proposed EG-CNN framework provides a highly accurate, robust, and explainable solution for plant disease diagnosis. Its multimodal architecture and strong generalization capability make it a promising tool for precision agriculture and real-time disease monitoring applications.
This study presents a deep learning-based framework for automated tomato leaf disease detection using a transfer learning approach built on ResNet18 architecture. The system is designed to classify 10 disease categories using image inputs resized to 224×224 pixels, leveraging ImageNet pre-trained weights to enhance feature extraction. The dataset consists of approximately 18,345 training samples with a batch size of 16 and trained over 10 epochs using the Adam optimizer and CrossEntropy loss function. Experimental results demonstrate strong classification performance, achieving a peak validation accuracy of 98.2% at epoch 9, with validation loss reduced to 0.64 from an initial value of approximately 1.8. The model shows high confidence predictions, with over 90% of test samples falling within the 90–100% confidence range. Per-class F1-scores range between 0.82 and 0.95, indicating consistent performance across multiple disease categories. The confusion matrix reveals strong diagonal dominance, confirming correct classification for most classes; however, specific misclassification patterns were observed. For instance, Bacterial_spot and Early_blight exhibit mutual confusion, while Septoria_leaf_spot shows complete misclassification into Tomato_mold, indicating limitations in distinguishing visually similar disease patterns. A balanced test dataset with class distribution ranging between 9% and 11% ensures unbiased evaluation. Additionally, Grad-CAM visualization confirms that the model focuses on biologically relevant regions such as lesion areas and leaf textures, improving interpretability. Despite achieving high accuracy under controlled conditions, the model’s generalization to real-world environments remains a challenge. The study highlights the need for improved robustness against variations in lighting, background, and disease severity. Overall, the proposed system demonstrates strong potential for precision agriculture applications, particularly in mobile-based disease diagnosis systems for farmers.