Research On Agricultural Pest Object Detection Model Based On YOLOv5

China is an agricultural country,so agricultural pests control becomes particularly important.Pest control is an important strategy to improve crop yield and quality.It can accurately and rapidly detect pests,which is of great significance to agricultural production.Traditional detection methods in the past mainly rely on expert experience and cost too much.In recent years,with the proposal of the concept of deep learning,various intelligent detection algorithms emerge one after another.Among all kinds of algorithms,the one-stage target detection algorithm is widely used because of its simple network structure,fast speed and high precision detection performance.In the field of agricultural pests,the detection effect of agricultural pests is not ideal due to the small data sets,dark and complex background,noise information and feature map loss and other problems.Aiming at the problems of the current one-stage detection algorithm in pest detection,this paper combined the target detection algorithm with the agricultural pest scenario,and made relevant improvements to the target detection problem in the agricultural pest scenario based on the first-stage target detection algorithm of YOLOv5.The main research work of this paper is as follows:1.Aiming at the problem of low detection speed caused by the large size of YOLOv5 algorithm network model.A lightweight YOLOv5 detection model was designed to improve the detection speed.First of all,the neck layer of the common convolutional module replaced by Ghost convolutional module,reduce the amount of calculation,speed up the detection speed;Then,in order to ease the detection accuracy,CBAM attention mechanism module is integrated.The experimental results show that compared with the original YOLOv5 model,the model size of the improved model is reduced by 45%,the detection time is shortened by 5%,and the balance between lightweight and detection accuracy is achieved.2.A new model was designed for the low performance of the lightweight model in the agricultural pest scene with complex background and excessive noise information.Firstly,CA attention mechanism is introduced to reconstruct the feature weights of target and noise information.Then adjust the receptive field module to better obtain the surrounding information of small targets;Finally,in order to solve the problem that convolution operation will cause the loss of deep feature map,a context feature fusion module is designed to take the information extracted from deep layer as the auxiliary information of small target pests,and finally improve the accuracy of pest detection.The experimental results show that the improved m AP reaches 90.7%,which improves the detection accuracy.Figure[33] Table[5] Reference[66]...