Plant Leaf Health Classifier Using Explainable AI

Tan, Kean Chan (2025) Plant Leaf Health Classifier Using Explainable AI. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Tan Kean Chan - FULL TEXT.pdf Restricted to Registered users only Download (5MB)

Abstract

This project focuses on developing an advanced plant leaf disease detection system by integrating deep learning models with Explainable AI (XAI) techniques. The project is designed to accurately identify diseases in plant leaves using image data of potato and tomato plants from the Plant Village dataset. These images were categorized into healthy and diseased classes for training and evaluation purposes. Several CNN architectures were employed, including EfficientNet, EfficientNetV2, and MobileNet. To optimize performance, hyperparameter tuning was conducted by experimenting with different optimizers, learning rates, and training epochs. Among all models, EfficientNetV2B3 achieved the best performance when trained with the Stochastic Gradient Descent (SGD) optimizer at a learning rate of 0.01 for 200 epochs. This configuration resulted in training accuracy of 99.78 % and validation accuracy of 97.79 %, with low training and validation losses. It also achieved strong and consistent classification results, with average precision, recall, and F1 score all at 0.97. To enhance transparency and interpretability, XAI techniques such as Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) were incorporated. These tools helped visualize the specific regions of leaf images that influenced the model’s predictions. LIME, in particular, proved highly practical for agricultural use due to its simple and intuitive explanations. This project contributes to sustainable farming by enabling early and accurate plant disease detection, empowering farmers to make timely and informed decisions. It represents a meaningful step forward in the application of AI for smart agriculture and food security.