Deep Learning Based Mobile Application for Automated Plant Disease Detection

Plant diseases remain a significant threat to global agriculture, necessitating rapid and accurate detection to minimize crop loss. This paper presents a lightweight, end-to-end system for plant leaf disease detection and severity estimation, optimized for real-time field deployment. We propose a custom Convolutional Neural Network (CNN), built using PyTorch, trained on the PlantVillage dataset to classify leaves as healthy or diseased with a test accuracy of 92.06%. To enhance its practical relevance, we incorporate a classical image processing pipeline using OpenCV and NumPy to estimate the severity of infection by computing the ratio of diseased to total leaf area. These capabilities are integrated into a cross-platform mobile application developed using React Native, with inference served via a Flask-based backend API. The mobile app enables users to capture or upload images and instantly receive diagnostic results, and severity percentages. Our system bridges the gap between deep learning research and real-world agricultural application by combining accurate classification, interpretable severity estimation, and mobile accessibility. This approach offers farmers a powerful, on-device digital assistant to monitor crop health and make informed intervention decisions. Experimental results demonstrate strong generalization performance, visual alignment of model attention with infected regions, and real-time usability in field conditions.