CNN BASED PLANT DISEASE CLASSIFICATION WITH GENERATIVE AI ADVISORY SYSTEM

Abstract

Plant diseases cause substantial crop yield losses worldwide, threatening food security in agricultural-dependent economies. This paper presents an end-to-end Plant Disease Detection and AI Advisory System that combines convolutional neural network (CNN)-based image classification with a large language model (LLM) for context-aware advisory generation. A MobileNetV2 model trained via transfer learning achieves 98.9% validation accuracy across ten disease classes covering six major crops. A curated disease_facts.json knowledge base stores structured symptom, cause, and treatment information for each class. The Gemini 2.0 Flash API is prompted exclusively from this knowledge base, ensuring that advisory outputs remain factually grounded and free from hallucinated chemical recommendations. A confidence-aware decision gate (threshold τ = 0.60) filters uncertain predictions before activating the advisory pipeline. The fully interactive Streamlit application allows farmers and agronomists to upload leaf images, receive instant diagnoses, and query a conversational chatbot for safe, expert-aligned care guidance. The system demonstrates that separating CNN classification from LLM explanation yields both high diagnostic accuracy and trustworthy advisory content.