A Lightweight CNN Architecture to Identify Various Rice Plant Diseases in Bangladesh

Abstract

Rice diseases has been a major concern all over the world in recent years. Early and accurate prediction of disease can help the farmers in applying proper treatment on the plants thus protecting the crop from pesticides and improving the overall growth. Deep learning based image processing methods can be a great solution in identifying various rice plant diseases accurately and precisely. For this research, we have collected a total 12 different types of rice disease images. The images has been pre-processed and augmented using different algorithms. Along with different state of the art CNN architectures, a lightweight CNN architecture has been proposed for identifying various rice plant diseases. The experimental result shows that our proposed model can identify the rice plant diseases with a mean validation accuracy of 95.4%. Considering small parameter size, it is evident that our proposed CNN model performs significantly well in detecting various rice plant diseases accurately.