DEEP CONVOLUTIONAL NEURAL NETWORK FOR AUTOMATED CLASSIFICATION OF RICE LEAF DISEASES USING A PUBLICLY AVAILABLE DATASET

Abstract

Rice leaf diseases are a major issue for the whole world's food security because they lower crop harvest and grain yield. Detecting these diseases at the initial stage is significant for dealing with crops effectively. Traditional detection methods depend on physical observation. This process takes time and there can be mistakes. A convolutional neural network (CNN), a deep learning model, is used in this research paper on a publicly available Rice diseases images dataset at Kaggle to classify the rice leaf diseases. The dataset consists of 5932 images and 04 folders named as healthy leaves, brown leaf spot (BLS), bacterial leaf blight (BLB), and Rice leaf smut (RLS). Before training, every image was normalized and resized to (128 x 128) pixels. The model is a grouping of six convolutional layers with max pooling, Rectified linear unit activation, and dropout regularization. These are followed by fully connected layers using softmax classification. Training used the Adam optimizer with a learning rate of 0.001 for 50 epochs. Accuracy, recall, precision, and F1 score were used to measure performance. The proposed model attained a validation accuracy of 92.60%. This was higher than K Nearest Neighbors at 56.84% and Support Vector Machine at 61.47%. These results show that the model can properly detect different rice disease types under numerous field situations. This study proposes that deep learning can make a substantial contribution to precision farming by providing a consistent and efficient method for disease diagnosis. Anticipated steps will enhance the dataset, implement the model in its intended environment, and design lightweight versions appropriate for mobile and Internet of Things (IoT) applications.