基于改进的CA-BiFPN_YOLOv5s车辆检测算法  

A vehicle detection algorithm based on improved CA-BiFPN_YOLOv5s

作　　者：刘宇辰 Liu Yuchen(Hubei University of Education,Wuhan 430205,China)

机构地区：[1]湖北第二师范学院,湖北武汉430205

出　　处：《汽车知识》2025年第1期243-247,共5页AUTOMOTIVE KNOWLEDGE

摘　　要：针对目前车辆检测当中存在的误检和漏检的问题,提出一种CA-BiFPN_YOLOv5s车辆检测算法:首先,在主干特征提取网络中加入协调注意力机制(CA)模块,使主干网络在特征提取时关注更重要的信息,从而提升目标检测精度;其次,采用加权双向特征金字塔网络(BiFPN)替换YOLOv5s网络中的PANet,加强模型提取多尺度特征的能力,提高了融合效率。实验表明:CA-BiFPN_YOLOv5s在BIT-Vehicle Dataset上车辆检测的平均精度均值(mAP)达到了94.8%,较YOLOv5s网络提高了2.8%,处理的帧率达到136.9 frame/s,满足实时车辆检测的要求。A CA-BiFPN_YOLOv5s vehicle detection algorithm is proposed to address the issues of false positives and false negatives in current vehicle detection.Firstly,a coordinated attention mechanism(CA)module is added to the backbone feature extraction network to focus on more important information during feature extraction,thereby improving the accuracy of object detection;By using BiFPN which is weighted replace the PANet network of YOLOv5s,the model's ability to extract multi-scale features is enhanced and fusion efficiency is improved.The experiment showed that CA BiFPN_YOLOv5s achieved a mean average precision(mAP)of 94.8%in vehicle detection on the BIT Vehicle Dataset,which is 2.8%higher than the YOLOv5s network.The processing frame rate reached 136.9 frame/s,meeting the requirements of real-time vehicle detection.

关 键 词：车辆检测 YOLOv5s 协调注意力机制 加权双向特征金字塔网络 

分 类 号：U472.9[机械工程—车辆工程]