Research And Application Of Insect Lightweight Detection Model Based On Convolutional Neural Network

The crops in China suffer serious losses due to diseases and insect pests every year.The problem of real-time monitoring of diseases and insect pests has been perplexing the staff.Traditional pest monitoring still requires workers to enter the field to observe the physical characteristics of pests and diseases,and visually detect and diagnose the pest situation in the area.This method has a large workload and low efficiency,and cannot predict the occurrence of diseases and insects in real time to meet the needs of current pest monitoring.With the development of deep learning,it is possible to use convolutional neural network technology to automatically predict farmland pests.In this paper,the existing detection model can only detect several species of insects,which does not conform to the actual scene.In order realize the recognition of multi-target insects,the data sets of 26 species of insects were made according to the insect pictures collected by the detection lamp.Then,a lightweight detection model based on YOLO-V3 was proposed.Finally,the intelligent detection lamp system of insect situation was built to realize the automatic monitoring of pests and diseases.The main research contents and results include:(1)Research on different insect detection algorithms based on convolutional neural Network.Target detection algorithms are mainly divided into two-stage target detection algorithms based on area generation network(R-CNN)and end-to-end one-stage target detection algorithms(YOLO).In this paper,two insect target detection algorithms based on Faster R-CNN and YOLO-V3 are trained respectively.Faster R-CNN adopted VGG-16 and Res Net-50 as the trunk characteristic network,and YOLO-V3 adopted CIOU as the loss function.The test results show that the YOLO-V3 insect detection model has higher accuracy.(2)Insect target detection algorithm based on depth separable convolution.This paper proposes a lightweight insect detection model,which is based on the idea of the classic YOLOV3 target detection algorithm.Firstly,lightweight Backbone(Light Net)is constructed for feature extraction through deep separable convolution,and then PANet is used for multi-scale feature fusion.Finally,three characteristic matrices of different sizes are output to predict pests of different sizes.CIOU is used as the loss function of regression prediction,so as better reflect the relative position of prior frame and real frame.Through lightweight algorithm are simulated with other contrast experiment,the results show that the proposed farmland pests recognition algorithm accuracy is the highest,at 93.73%,and the model of lightweight,can be deployed on work force lower equipment,reduce the cost of the equipment,accurately predict the status of the plant diseases and insect pests in farmland,in practice show that The algorithm can effectively solve the problems of large number of pests,pest accumulation,background interference,and has strong robustness.(3)Design and implementation of lightweight insect target detection system.The system helps workers predict pest infestation in an area and protect local crops.Pyqt is used to build the client,and the detection results are sent to the gateway and server using FTP and HTTP protocols.The communication between the single chip microcomputer and the client is realized through RS232 serial port,so as complete the automatic insect disaster warning.In this paper,a lightweight insect target detection model is constructed to realize automatic detection of 26 species of insects,reducing the problem of missing detection in the complex background,which can be used for actual pest monitoring.