Research On Field Corn Pest Identification And Monitoring System Based On Deep Learning

China as a big corn producer,a large number of corn pests invade corn in the growth process of corn every year,resulting in a large-scale reduction of yield and quality of corn,which can lead to a sudden harvest in serious cases.Therefore,timely and accurate identification of corn pests is an important prerequisite for effective control of corn pests.With the great progress of deep learning in the field of identification and detection,it has brought a new solution for corn pest identification.This study analyzes the pests in corn field,makes the data set of common corn field pests,compares the mainstream target detection model through theoretical analysis and experiment,and selects the initial target detection model of this study.At the same time,according to the characteristics of corn pests and the structural characteristics of the initial target detection model,the initial target detection model is optimized and improved,which is supported by theoretical analysis and relevant experiments.Finally,based on the improved target detection model of corn pests,a corn pest identification and monitoring system is studied and designed.The main contents of the study include:(1)By integrating the image data of corn pests in the existing literature and the image data of corn pests searched on the Internet,after manually screening and removing the wrong and unusable image data of corn pests,10 kinds of common pests in corn field were sorted out,including 3514 image samples.In order to ensure the robustness and generalization ability of the corn pest detection model,common image enhancement methods are used to block,zoom,translate,rotate,change brightness,add noise and mirror image to the original image,and the image data of corn pests sorted in the early stage are expanded.After the expansion,the number of image data of corn pests reaches 9492.Finally,the open source software Label Img is used to label the image data of corn pests,Complete the production of field corn pest data set.(2)Through the research and analysis of the structural characteristics and training process of one-stage and two-stage target detection models,after building the deep learning framework and deploying the corresponding target detection model,the self-built corn pest data set is used for training and testing.Combined with the experimental results,model characteristics and actual needs,the Faster R-CNN two-stage target detection model is selected as the initial model of this study.(3)Through the analysis and research of the initial model feature extraction network,Res Net50 is used to replace the VGG16 feature extraction network of the original model to improve the feature extraction ability of the model.At the same time,through the research and analysis of clustering algorithm,K-means++ clustering algorithm is selected,and the new clustering distance based on intersection union ratio is used to cluster the frame selection information of corn pest object instead of the original European distance,so as to obtain the distribution of frame selection information of corn pest object.According to the clustering results,the increase ratio is 2:3;3:2,delete the 512*512 anchor frame size that does not meet the corn pest frame selection,and increase the 64*64 anchor frame size.At the same time,add attention mechanism to Res Net50 feature extraction network,strengthen useful feature information and suppress useless feature information,so as to further enhance the feature extraction ability of the network and finally improve the performance of the model.(4)In order to use the improved corn pest detection model to solve practical problems,a corn pest identification and monitoring system was designed.Based on the improved corn pest detection model,a corn pest identification and monitoring system is built to realize the functions of uploading,reading,detection and analysis of pest images.