Early Plant Stress Detection using Thermal Leaf Patterns with TAP-EfficientNet for Precision Agriculture

Abstract

The objective of this study is to employ the proposed TAP-EfficientNet model for the early detection of plant stress using thermal leaf patterns, aiming to improve diagnostic accuracy and computational efficiency in precision agriculture. Group 1 is the standard EfficientNet baseline model. Group 2 is the proposed TAP-EfficientNet model. A sample size of 500 thermal leaf images is used for each group, and data is collected across various time intervals and stress conditions (e.g., water deficit, disease). The models' classification accuracy, precision, recall, F1-score, and inference delay are all calculated. The output demonstrated that the TAP-EfficientNet model has better classification results than the standard EfficientNet model in terms of 5.4% higher accuracy, 4.8% higher precision, 6.2% higher F1-score, and [e.g., 12.5%] lower inference delay. The results of the experiment indicate that the suggested TAP-EfficientNet model can detect early plant stress more effectively than the standard EfficientNet model, making it highly suitable for real-time monitoring and deployment in precision agriculture.

Code and data availability

The paper uses a Kaggle-sourced thermal leaf image dataset and a custom TAP-EfficientNet model, but provides no public URL, deposit, or availability statement for the dataset, code, or trained model. The Kaggle reference is generic (cited via a PlantVillage paper) and no authors' repository is given, so no paper-quali­

No evidence-backed public reproduction asset is currently recorded.