Research On Leaf Disease And Pest Identification Method Of Watermelon And Melon Based On Convolutional Neural Network

Watermelon and melon are important categories of fruits for daily consumption in China,with annual production and consumption ranking among the top in the world for a long time.In view of the huge population base of China brings watermelon consumption demand,it is necessary to ensure that the domestic production of watermelon and the people’s food demand are met.Pests and diseases have been one of the important factors affecting crop yield in agricultural production,and is an important research direction in agricultural production technology.In this paper,we focus on five common pests and diseases in the growth of watermelon plants,and realize the rapid detection of five common pests and diseases in watermelon through machine vision technology,which provides an effective means of pest and disease detection for watermelon pest and disease control.The main research contents and results of this paper are as follows:(1)Data collection was performed by smart phones and Io T devices for diseased leaves of watermelon in greenhouse environment,and 5871 images of watermelon leaves containing four kinds of pests and diseases were obtained,and the own dataset was established by this.(2)For the two practical problems of complex environment in greenhouse and arithmetic limitation of hardware equipment,the original YOLO v5 could not meet the working needs,so this paper improved the backbone of YOLO v5 by using Shuffle Net v2 and made targeted improvements to the reconstructed model,and finally achieved better detection results.(3)In this paper,we use the channel pruning method to compress the size of the improved convolutional neural network model,Pruned-YOLO v5+Shuffle(PYSS)network model,to 1.3MB,while maintaining 98% of the recognition precision of the original model.The compressed model is more suitable for deployment on edge devices,which makes the research method more applicable to a wide range of platforms.(4)Through the effective combination of Io T devices and edge computing devices,the data collection of pests and diseases of watermelon,the regular training of neural network models,the efficient and fast database update and model iteration can ensure the long-term availability of the models.The research in this paper can quickly and effectively realize the high-precision identification and detection of watermelon leaf pests and diseases,which provides a technical guarantee for the rapid response of watermelon pest and disease control and facilitates the subsequent staff to provide a theoretical basis and technical support for the accurate and effective disposal of watermelon pests and diseases.