基于注意力机制改进YOLOv7算法的车辆识别检测  

Vehicle recognition and detection based on improved YOLOv7algorithm using attention mechanism

作　　者：牟俊宇 陈菲 韩钰松 刘超[1] 白云贵 刘丽霞[1] MOU Junyu;CHEN Fei;HAN Yusong;LIU Chao;BAI Yungui;LIU Lixia(Faculty of Mathematics and Artificial Intelligence,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China)

机构地区：[1]齐鲁工业大学(山东省科学院)数学与人工智能学部,山东济南250353

出　　处：《齐鲁工业大学学报》2025年第2期8-16,共9页Journal of Qilu University of Technology

基　　金：2022年度校级人才培养和教学改革项目(242210030115),科教产融合试点工程基础研究类项目(2023PX034)。

摘　　要：实时对象目标检测是计算机视觉中非常重要的主题,是计算机视觉系统中的必要组件。针对城市交通汽车检测问题,可以利用YOLO模型实现对道路车辆检测的智能化。为优化计算机对车辆的实时检测能力,该研究提出一种基于注意力机制的改进YOLO算法的车辆识别检测算法,使用YOLOv7为主体,在YOLOv7网络模型的Backbone和Head模块引入注意力机制,以适应不同车辆的识别任务。在Roboflow的公开数据集上进行实验,结果表明,改进后的网络模型相较于原始的YOLOv7网络模型,汽车漏检情况得到改善,在同一数据集下相比YOLOv7网络模型提升了0.9%,P mA值达72.2%,检测效果可基本满足汽车检测应用需求。Real time object detection is a very important topic in computer vision and a necessary component in computer vision systems.For the problem of urban traffic vehicle detection,the YOLO model can be used to achieve intelligent road vehicle detection.To optimize the real-time detection capability of computers for vehicles,this paper proposes an improved YOLO vehicle recognition and detection algorithm based on attention mechanism.YOLOv7 is used as the main body,and attention mechanism is introduced for the Backbone and Head modules of the YOLOv7 network model to adapt to different vehicle recognition tasks.Experimental validation is conducted using a publicly available dataset of Roboflow,demonstrating notable improvements in car detection performance compared to the original YOLOv7 model.The enhanced network achieves a P mA value of 72.2%,marking a 0.9 percentage point increase over the baseline YOLOv7 model on the same dataset.These results underscore the efficacy of the proposed algorithm in meeting the requirements of vehicle detection applications.

关 键 词：深度学习 多目标检测 YOLOv7模型 车辆识别 注意力机制 

分 类 号：U495[交通运输工程—交通运输规划与管理]