Efficient net-based deep learning model for accurate plant disease classification and diagnosis

Abstract

Diseases are a major drawback to crop production, productivity, and food security in nations affected by plant diseases. This work proposes an efficient framework for the automated recognition and diagnosis of diseases within plants using a convolutional neural network architecture known as EfficientNet. For this study, a large dataset containing sharp images of impaired and hale plant organs belonging to various species was collected. Common data preprocessing steps such as Resizing and Augmentation were used to reduce overfitting and increase the model’s ability to generalize. Finally, EfficientNet was trained for multi-class disease segmentation with the validation accuracy of 95%. The model showed high value of accuracy and recall and solved problems of the differentiation of visually similar diseases among the different categories. It is so from the following view: These results show the possibility of this approach as the practical tool for early disease detection and management in agriculture on large scale. Further studies are going to be conducted enlarging the data set, enhancing the transferability of the developed model, and examining how the app is best to be disseminated, for instance, via mobile applications or Internet of Things (IoT) devices for constant farming inspection.

Code and data availability

The paper uses a Kaggle-sourced plant disease image dataset and an EfficientNet model, but provides no authors' public code, trained model, or dataset deposit; the dataset is a cited third-party resource, not a paper-specific asset.

No evidence-backed public reproduction asset is currently recorded.