Rice Disease Detection And Recognition Based On Machine Vision

At present,the research on automatic detection and recognition of rice diseases based on machine vision technology has been widely recognized and valued.In this paper,the images of rice blast and bacterial blight diseases in indoor and field backgrounds are used as the research objects to study the automatic detection and recognition methods of rice diseases.The main work is as follows:(1)A high-quality,high-performance,easy-to-operate digital SLR camera Model-Canon EOS 5D Mark IV was selected as the image acquisition device,and a total of 2094 pictures of rice diseases were taken.First,pre-process the pictures,establish a rice disease database according to different shooting environments and disease types,and use CVAT annotation software to mark the disease spots in the pictures.Two 50 pictures of rice blast and bacterial blight disease in different backgrounds were selected as the test set of traditional machine vision methods.In addition,800 field background disease pictures were selected as the deep learning training set,and these pictures were enhanced by noise,blur,color,and mirror transformation methods.The expanded data set contains a total of 3200 pictures,and according to Mask R-CNN The data format is required,and the COCO data set of field rice diseases has been produced.(2)Through theoretical analysis and overview,this paper selects Grab Cut,a semiautomatic segmentation algorithm based on human-computer interaction,to segment rice disease images.The Grab Cut rice disease recognition software was developed based on Qt’s graphical user interface technology.It used 200 disease test set images under different backgrounds for evaluation.The results showed that the indoor disease image segmentation IOU value can reach more than 95%,and the error segmentation rate is E.The value is less than 1%,and the field disease image segmentation IOU value is less than 80%,and the error segmentation rate E value reaches more than 7%.The Grab Cut algorithm must use manual selection of feature areas during the application process,resulting in a low degree of automation.Through comparative analysis,it is concluded that the algorithm is only suitable for image recognition with relatively simple backgrounds,and cannot be widely used in complex backgrounds.(3)In response to the problems found in the Grab Cut algorithm,a target detection algorithm based on deep learning,Mask R-CNN,is proposed.The Mask R-CNN model was constructed based on the Tensor Flow environment,and the feature extraction was replaced by convolutional neural networks.The field rice disease data set was trained to achieve effective detection and identification of two kinds of disease spots of rice blast and bacterial blight in a complex background.Continue to adjust the model training times again and again to view the changes in network function loss trends,understand the training effect models,using 200 test set to evaluate pictures showed that rice blast and bacterial leaf blight recognition rate reached82.5% and 90.7% respectively,F-measures reached 88% and 94.2% respectively.Compared with the traditional machine vision Grab Cut method,the IOU values of the two kinds of disease recognition have increased by about 10%.The use of machine vision technology to accurately identify rice diseases has important theoretical and practical significance for ensuring food security,increasing farmers’ income,improving the quality of agricultural products,and achieving sustainable agricultural development.It also lays the foundation for the research of rice disease identification technology.