Rice Disease Prediction using Deep Learning for farmers

This paper addresses the problem of classifying diseases prevalent in Rice (Paddy) crops in Karnataka by leveraging deep learning and feature extraction techniques. Two distinct methodologies were explored to enhance disease identification accuracy. The first method utilized a transfer learning approach with ResNet-50, incorporating dynamic learning rate scheduling with warm-up and cosine annealing strategies. This ensured efficient convergence while maintaining high generalization performance. The second method combined deep learning-based feature extraction with handcrafted features such as HOG, LBP, and color histograms. This hybrid feature fusion approach enabled a comprehensive representation of disease patterns. Experimental results demonstrated that both methodologies achieved high classification accuracy, with the hybrid model excelling in complex disease identification scenarios.