Plant Disease Classification System with Explainable AI
Deep learning classification system achieving 91% accuracy via model stacking + CNNs, with Grad-CAM explainability and LLM-powered treatment recommendations.
Problem
Plant disease identification requires expert knowledge that is inaccessible to most farmers. Existing CV models lack explainability and actionable treatment guidance, limiting practical adoption.
Action
Designed and trained multiple CNN models for five-class disease image classification, improved accuracy to 91% via model stacking strategy, and addressed class imbalance through data augmentation and resampling. Applied Grad-CAM for decision process visualization. Developed a Streamlit web application integrated with ChatGPT for automated diagnosis and treatment recommendations.
Result
Achieved 91% classification accuracy across 5 disease categories. Grad-CAM visualizations enhanced model transparency. LLM integration provided actionable treatment plans, forming a complete decision support system deployable via smartphone.
Learnings
Learned the importance of combining classification accuracy with explainability for real-world trust, and how to integrate LLMs to bridge model predictions and human-actionable guidance.