Using Deep Learning Techniques to Detect Rice Diseases from Images of Rice Fields

Abstract

Rice is a staple food feeding more than half of the world’s population. Rice disease is one of the major problems affecting rice production. Machine Vision Technology has been used to help develop agricultural production, both in terms of quality and quantity. In this study, convolutional neural network (CNN) was applied to detect and identify diseases in images. We studied 6 varieties of major rice diseases, including blast, bacterial leaf blight, brown spot, narrow brown spot, bacterial leaf streak and rice ragged stunt virus disease. Our studied used well known pre-trained models namely Faster R-CNN, RetinaNet, YOLOv3 and Mask RCNN, and compared their detection performance. The database of rice diseases used in our study contained photographs of rice leaves taken from fields of planting areas. The images were taken under natural uncontrolled environment. We conducted experiments to train and test each model using a total of 6,330 images. The experimental results showed that YOLOv3 provided the best performance in term of mean average precision (mAP) at 79.19% in the detection and classification of rice leaf diseases. The precision obtained from Mask R-CNN, Faster R-CNN, and RetinaNet was at 75.92%, 70.96%, and 36.11%, respectively.