Insect Detection and Counting Based on YOLOv3 Model

Insect detection is an important means to predict the detection of agricultural pests. As deep learning tends to mature, deep learning is widely used in agriculture, including insect detection, but recognition can only operate on a single insect, it does not conform to the actual scene. In order to realize the recognition of multi-target insects and solve the problem of difficulty in recognition caused by background problems, this paper adopts the target detection network based on YOLOv3 model to realize the detection of multiple targets, and apply CSPDarkenet-53 as the backbone feature extraction network. It solves the problem that the background is difficult to extract features, and improves the model. Further, the CIOU is used as the loss function of regression prediction, which can reflect the relative position of the priori box and the real box more effectively. The priori box of the data set is obtained by using the K-means clustering algorithm. The results show that the accuracy of the improved YOLOv3 model can be achieved 90.62%. This model improves the accuracy of 3% compared with the yolov3 model, and verifies the feasibility of object detection based on deep learning in insect detection.