DETECTION OF RICE LEAF DISEASES USING CONVOLUTIONAL NEURAL NETWORK

Abstract

Purpose: The objective of this paper is to provide a system that helps the farmers to identify the disease in the rice leaf using its image.

Methodology: The dataset is first assembled by gathering unhealthy rice leaf pictures. The data is then preprocessed and expanded through a number of data augmentation techniques. The data is then split into two sections: training and testing dataset. It is then classified by the CNN model. Here transfer leaning is used using a pre-trained model named Inception v3. Finally, the name of the disease and its remedy is displayed to the user.  

Findings: For image classification, a deep learning model is trained with labelled images in order to learn how to identify and classify them according to visual patterns. We used an opensource implementation comprising CNN, named Inceptionv3, which is provided as part of the Keras module and it was recognized as validation on ImageNet. For each input image, the feature vector has size features = 2048. For this module, the size of the input image is fixed to height x width = 299 x 299 pixels. Once the convergence was achieved after a few iterations, the batch size was increased to 64 images and the number of epochs to ten. The Adam optimizer was utilized with a 0.001 learning rate and category cross entropy loss. Then, for the identification of diseases in the rice leaf, Transfer Learning is applied using a pre-trained Inception v3 model. For diagnosing the infection in the rice leaf, an accuracy of 94.48 percent was reached. The name of the disease is displayed for the given input image along with the recommended solutions by the developed model.

Originality/value: The training accuracy acquired during training is 96.34 percent, and the validation accuracy is 94.48 percent. Thus, the adjusted inception v3 model may be utilized as a diagnostic technique to identify the disease in the rice leaf and give necessary expertise recommendations to treat the diseases.