一种基于YOLOv5s-Lite车辆目标识别方法  

A Vehicle Target Recognition Method Based on YOLOv5s-Lite

作　　者：李孟歆[1] 李志秀 姜政 LI Meng-xin;LI Zhi-xiu;JIANG Zheng(School of Electrical and Control Engineering,Shenyang Architectural University,Shenyang Liaoning 110000,China)

机构地区：[1]沈阳建筑大学电气与控制工程学院,辽宁沈阳110000

出　　处：《计算机仿真》2025年第1期149-154,共6页Computer Simulation

基　　金：国家自然科学基金项目(62133014);辽宁省自然科学基金(20180550286)。

摘　　要：针对现有识别算法对车辆目标识别效果欠佳,存在误检漏检等问题,提出一种基于改进YOLOv5s算法的车辆识别方法。首先,提出采用ShuffleNetV2网络替代YOLOv5s的主干网络,实现网络轻量化设计;其次将原网络CIoU损失函数修改为EIoU损失函数,提高对车辆目标识别能力;然后采用DIoU-NMS替代原网络的NMS后处理方法,降低网络的漏检率;最后采用K-means++聚类算法重新生成聚类锚框。通过在晴天与阴天两种场景下对比可知,所提算法相比于原算法,模型大小减小35.7%,参数量减少54.7%,识别速度提升了46.31FPS,满足实时性需求,同时提高了车辆识别精度。To address the poor performance of existing recognition algorithms for vehicle target identification,which suffer from issues such as false positives and missed detections,a vehicle recognition method based on an improved YOLOv5s algorithm is proposed.Firstly,the ShuffleNetV2 network is proposed to replace the backbone network of YOLOv5s to achieve a lightweight network design.Secondly,the CloU loss function of the original network is modified to the EIoU loss function to improve the ability of vehicle target recognition.Then,DIoU-NMS is used to replace the original post-NMS processing method to reduce the rate of missed detection.Finally,the K-Means++clustering algorithm is used to regenerate the cluster anchor frame.By comparing the two scenarios on sunny and cloudy days,it can be seen that compared with the original algorithm,the proposed algorithm reduces the model size by 35.7%,the number of parameters by 54.7%,and the recognition speed by 46.31FPS,which meets the real-time requirements and improves the vehicle recognition accuracy.

关 键 词：轻量化 损失函数 聚类锚框 深度学习 车辆识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]