A Novel Deep Learning Based Model for Classification of Rice Leaf Diseases

Abstract

Rice is the primary source of food for a vast population worldwide, especially for most Asian countries. Diseases in rice leaves can have disastrous outcomes and cause massive losses in the agricultural sector. Thus, there is a need for early automatic detection of rice leaf diseases. Many methods have been proposed before in order to solve this task which involves the use of deep learning because of its good results. In this work, a novel transfer learning-based model has been suggested for the automatic classification of 5 different classes of diseases. DenseNet 201 has been used as the base model with weights from ImageNet. Instead of assigning random weights, the weights from the pre-trained network have been set but the layers have been trained from scratch on the given dataset in order to produce results. The proposed deep learning-based model shows better performance than the other existing state-of-the-art algorithms by achieving the training accuracy of 97.04 % and an accuracy of 95.44 % on the validation dataset respectively. Although the dataset has noises present and no effective preprocessing steps were done, the model performed quite well. This work provides a new method for deep learning-based classification of rice diseases.