Optimized CNN for Chili Leaf Disease Classification

This technical report presents an optimized Convolutional Neural Network (CNN) approach for the classification of chili leaf diseases, addressing the challenges of traditional manual inspection methods. The study evaluates various CNN architectures, including MobileNet, ShuffleNet, ResNet, and VGG16, to identify the most effective model for disease classification. The findings indicate that RegNetX-148MF achieves superior classification performance with a validation accuracy of 97.65%, while also being computationally efficient. The report details the application of pruning techniques to reduce unnecessary parameters and floating-point operations, enhancing the model's efficiency for deployment on low-resource devices. Additionally, damping-based fine-tuning is employed to improve model stability and accelerate convergence without increasing complexity. The research emphasizes the potential of lightweight CNN architectures for real-time agricultural disease detection, providing a practical solution for farmers to efficiently diagnose plant diseases and improve crop health management. This approach contributes to sustainable farming practices and food security.