XSE-TomatoNet: An explainable AI based tomato leaf disease classification method using EfficientNetB0 with...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Accurate tomato leaf disease diagnosis is vital for agricultural productivity and preventing ecosystem harm.
• Deep learning models show promise but struggle with subtle disease variations.
• Existing methods require improvement for precise and efficient disease identification.

Purpose of the Study:

• To introduce XSE-TomatoNet, an enhanced deep learning model for accurate tomato leaf disease classification.
• To improve upon existing EfficientNetB0 architecture by incorporating Squeeze-and-Excitation (SE) blocks and multi-scale feature fusion.
• To evaluate the performance and interpretability of XSE-TomatoNet for practical application in agriculture.

Main Methods:

• Developed XSE-TomatoNet by enhancing EfficientNetB0 with Squeeze-and-Excitation (SE) blocks and multi-scale feature fusion.
• Extracted multi-scale features, refined them with SE blocks, and merged them using Global Average Pooling.
• Employed data augmentation, ablation studies, 10-fold cross-validation, LIME, SHAP, Grad-CAM, and Grad-CAM++ for evaluation and interpretability.

Main Results:

• XSE-TomatoNet achieved a high accuracy of 99.11% (99% precision and recall), significantly outperforming MobileNet (87.44%) and VGG-19 (95.50%).
• 10-fold cross-validation demonstrated strong generalization with an average training accuracy of 99.41% and validation accuracy of 98.88%.
• Model interpretability techniques (LIME, SHAP, Grad-CAM, Grad-CAM++) provided insights into decision-making and visual disease identification.

Conclusions:

• XSE-TomatoNet offers a highly accurate and robust solution for tomato leaf disease classification.
• The enhanced model architecture effectively captures subtle disease variations, improving diagnostic precision.
• Integration into a web-based system facilitates practical adoption by tomato cultivators, enhancing crop management.