Rice Leaf Disease Detection with Transfer Learning Approach

Abstract

Rice (Oryza sativa) is among the most widely cul-tivated crops all over the world. The seed of the grass species Oryza sativa is commonly identified as rice. Rice is consumed all over the world as a main source of carbohydrate, specially in Asian countries. As a South Asian country, our homeland Bangladesh has identified rice as its staple food. Throughout the world, rice leaf diseases cause a huge loss in rice production each year. Traditionally, rice leaf diseases are detected in laboratory tests, which is time consuming. If machine learning and computer vision based approaches-which are faster and more accurate comparing to manual detection of rice leaf diseases- can be implemented to detect rice diseases, a substantial amount of production loss pertaining to these diseases can be mitigated. Deep learning frameworks, such as, convolutional neural networks (CNN) shows higher efficacy in image classification and object detection from images. They can be utilized to classify various rice diseases and, as a result, can play an important role in early detection of rice diseases and, consequently, improving the production. In this paper, we have utilized transfer learning approach by using three pretrained CNN models: InceptionV3, DenseNet201, and EfficientNet V2S to detect five prominent diseases of rice (Oryza Sativa) leaves along with healthy leaves seen in our country and have demonstrated extensive comparison between these models. Among the models, DenseNet201 showcased the highest accuracy which was 92.05%.