Rice disease detection for sustainable farming using machine learning

Abstract

The rice leaf diseases cause not only loss in their yield, but they are also a significant source of food insecurity especially in countries that are dependent on agriculture like Bangladesh where rice is the staple crop. The reason is that such diseases should be identified early enough and properly in order to take care of the crops and make relevant decisions. This paper provides one of the deep learning models utilizing image, a model that includes the use of digital images to automatically identify and categorize rice leaf diseases. The dataset of 1560 rice leaf images divided into six classes, five disease classes and one healthy class of images were gathered through the assistance of expert validated samples that were collected at Bangladesh Rice Research Institute (BRRI) and through the addition of publicly available internet access to online farming image collections in order to enhance the dataset variety. Preprocessing of the photos included resizing, normalization, and data augmentation and was followed by the division of the photos into training, validation, and testing sets in 70:20:10. A convolutional neural network was designed and trained in an original way and other known convolutional neural networks were used as benchmark models under the same experimental conditions. The test analysis indicates that the proposed model has had a total test performance of approximately 91.67% and the training and validation performance was approximately 94.87% that was the indication of persistent learning behaviour and consistent generalisation performance.The results demonstrate that the proposed method works well in the classification of rice leaf diseases and can be useful in practice with regard to the application in the agricultural monitoring and decision-support systems.