BotanicX-AI: Identification of Tomato Leaf Diseases Using an Explanation-Driven Deep-Learning Model

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Early and accurate detection of tomato diseases is vital for minimizing crop losses.
• Tomato cultivation faces significant threats from various infectious diseases and pests.
• Accessible leaf image analysis can empower farmers with timely disease management strategies.

Purpose of the Study:

• To develop and evaluate a deep learning model for identifying nine distinct tomato leaf diseases and healthy samples.
• To assess the performance of the EfficientNetB5 architecture on a tomato leaf disease dataset.
• To enhance model interpretability for practical agricultural applications.

Main Methods:

• Implementation of the EfficientNetB5 model utilizing a tomato leaf disease (TLD) dataset.
• Direct image classification without employing segmentation techniques.
• Application of gradient-weighted class activation mapping (GradCAM) and local interpretable model-agnostic explanations for interpretability.

Main Results:

• The EfficientNetB5 model achieved high average accuracies: 99.84% (training), 98.28% (validation), and 99.07% (testing) over 10 cross-folds.
• The model demonstrated robust performance in visually distinguishing between healthy and diseased tomato leaves.
• Interpretability methods provided insights into the model's decision-making process.

Conclusions:

• The study successfully demonstrates the efficacy of EfficientNetB5 for automated tomato disease identification using leaf images.
• High accuracy and model interpretability are key for integrating AI tools into agricultural practices.
• This approach offers a promising solution for early disease detection, contributing to sustainable tomato farming.