文献类型：会议论文

英文题名：Road traffic target detection based on improved YOLOv5 algorithm

作者：Huang, Junchao[1]; Qi, Chengming[1]

第一作者：Huang, Junchao

机构：[1] College of Urban Rail Transit and Logistics, Beijing Union University, Beijing, 100101, China

第一机构：北京联合大学城市轨道交通与物流学院

通讯机构：[1]College of Urban Rail Transit and Logistics, Beijing Union University, Beijing, 100101, China|[1141751]北京联合大学城市轨道交通与物流学院;[11417]北京联合大学;

会议论文集：International Conference on Automation Control, Algorithm, and Intelligent Bionics, ACAIB 2024

会议日期：June 7, 2024 - June 9, 2024

会议地点：Yinchuan, China

语种：英文

外文关键词：Intelligent transport; lightweight; model pruning; YOLO

摘要：As economic development and social progress bring convenience, they also bring a large number of problems. In the field of transport, the increase in the number of vehicles travelling on the road has led to an increase in traffic problems. With the introduction of intelligent transport, the problem of road traffic has been alleviated. in the field of traffic how to quickly and accurately identify the surrounding information to assist the driver in make decisions is crucial. In this paper, an improved YOLO algorithm model is proposed for solving related problems. In order to reduce the number of parameters in the model and accelerate the inference speed of the model, it is proposed to use Faster-Block as the backbone network, and further reduce the number of parameters by using GSConv convolution in the Neck part, and finally, the model is further compressed by using a pruning algorithm that learns convolution efficiently in a network thinning. And the improved algorithm was validated in the VOC dataset as well as the CCTSDB dataset for experiments, pruning the size of the model as well as the number of parameters while keeping the model mAP/0.5 without too much change, which reduces the inference time and achieves good results, and is more convenient to be used for deployment. ? 2024 SPIE.