DETECTION OF CROP LEAF DISEASES AND INSECT PESTSBASED ON IMPROVED FASTER R-CNN

摘要

Among the green and environmental protection methods of pest control, ecological control can not only reduce the land pollution caused using pesti- cides, but also ensure the green planting of crops. The ecological regulation and prevention of crop dis- eases and insect pests is not only a way to control diseases and insect pests, but also a way to protect the ecological system and rural ecological environ- ment. To protect the ecological environment of agri- culture and forestry, aiming at the problem that most of the existing methods are difficult to realize the rapid and high-precision recognition of the irregular areas of crop leaf diseases, a method of detecting crop leaf diseases based on Faster R-CNN is pro- posed. In this paper, the field crops under natural light as the research object, after the manual removal of redundant images, the collected crop and its leaf disease images were pre-processed. This includes cropping, normalization, image enhancement, and more to build universal data sets. This paper intro- duces the Faster R-CNN convolutional neural net- work into the detection of crop leaf diseases and in- sect pests, and improves the structure of Faster R-CNN. In this paper, residual network is used to ex- tract image features, and region proposal network uses sliding window mechanism to generate target candidate frames, which improves the model perfor- mance of Faster R-CNN. Based on Pytorch and the open source framework of MM detection, the exper- imental demonstration of the proposed method shows that the recognition accuracy and loss value of the proposed method are the best, and it is superior to other comparison methods, which can provide ref- erence for the accurate identification of crop leaf dis- eases.