Bean Leaf Disease Classification System

This project develops an efficient and accurate classification system for bean leaf diseases, leveraging advanced deep learning and optimization techniques. Its core motivation is to empower farmers with a robust, real-time diagnostic tool accessible on resource-limited devices like smartphones. 📱

Key Features

Disease Coverage 🌿

The system is trained to identify and classify three distinct bean leaf conditions:

• Healthy
• Angular Leaf Spot
• Bean Rust

Technical Architecture 🧠

• Model: Employs a Modified EfficientNetV2 model, designed for both lightweight performance and high effectiveness.
• Accuracy: Achieves exceptional performance, demonstrating up to 97.76% accuracy on test datasets.

Mobile Optimization 🚀

• Deployment Ready: Specifically optimized for TensorFlow Lite (TFLite) implementation, ensuring readiness for mobile deployment.
• Efficiency: Features a compact model size of only 6.18 MB and a rapid inference time of just 0.0594 seconds.

Interpretability 🔍

The system integrates Grad-CAM Visualization, providing critical interpretability. It highlights specific regions on the leaf image (lesions or spots) that the model used to make its prediction.

Associated Research 📚

This project is the official repository for the research paper: "Classification of Bean Leaf Lesions Using Modified EfficientNetV2 for Implementation in TensorFlow Lite." It also includes relevant presentation slides.