High-accuracy Object Detection Based on YOLOv3 Under Different Weather Conditions

Abstract

To improve the accuracy and speed of image target detection under complex weather conditions, an image target detection method based on the YOLOv3 algorithm and image enhancement processing is proposed. Firstly, we make a data set specifically for image target detection by filtering the VOC 2007 data set. Then we select the yolov3 target detection model as the infrastructure network of image target detection. We improved the network structure for better accuracy. Then, the original data is enhanced by setting different atmospheric and channel conditions, and we use the improved YOLOv3 model to detect the image target and calculate the accuracy. Experimental results show that the system has good prediction accuracy for all 20 categories in the original data set. In addition, the image object detection method based on the combination of the YOLOv3 algorithm and image enhancement processing can effectively complete the task of image object detection under complex weather conditions.