Published Paper Details:

The plant disease detection system is an intelligent, ai-powered platform designed to simplify the process of identifying and diagnosing plant diseases from leaf photographs using deep learning and computer vision techniques. in today's agricultural era, farmers and agronomists frequently deal with crop diseases that cause 20-40% annual yield loss globally, making early and accurate diagnosis critical for food security and sustainable agriculture. the proposed system addresses this challenge by integrating convolutional neural networks (CNN), transfer learning, and gradient-weighted class activation mapping (grad-cam) into a unified framework.the system allows users to upload a leaf photograph through a web interface, which is then processed through multiple stages including image preprocessing, normalization, data augmentation, and feature extraction using the mobilenetv2 deep learning architecture pre-trained on the imagenet dataset containing 1.2 million images. the extracted features enable accurate visual understanding of disease patterns, allowing efficient classification of plant diseases based on leaf texture, color, spot characteristics, and lesion patterns. the classified result is then passed to a disease information module which generates accurate, concise, and context-aware responses including disease name, severity level, treatment recommendations, and prevention guidelines.the system is implemented using a modern full-stack architecture, with a flask-based backend for rest api processing, a drag-and-drop web interface for user interaction, and tensorflow with keras for deep learning model training and inference. the training pipeline follows a two-phase strategy -- transfer learning with frozen backbone layers followed by fine-tuning of the top 30 layers -- achieving 97.8% top-1 accuracy and 99.5% top-3 accuracy across 38 plant disease categories on a curated dataset containing 15,000 leaf images collected from 14 plant species including tomato, potato, apple, corn, grape, pepper, strawberry, peach, and cherry. additionally, grad-cam visualization ensures model explainability by highlighting the exact leaf regions that triggered the disease prediction. the plant disease detection system provides real-time inference under 10 milliseconds, a lightweight 14mb deployable model, and a user-friendly interface, making it suitable for agricultural, research, and field applications.