基于VOD-YOLOv5的车辆检测算法研究  

Research on Vehicle Detection Algorithm Based on VOD-YOLOv5

作　　者：沈正坤 王正超 尹怀仙[1] 焦博文 李慧 SHEN Zhengkun;WANG Zhengchao;YIN Huaixian;JIAO Bowen;LI Hui(College of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266071,China;Department of computer science,Qingdao financial vocational school,Qingdao 266100,China;State Grid Shandong Electric Power Company Changyi City Power Supply Company,Chang Yi 261300,China)

机构地区：[1]青岛大学机电工程学院,山东青岛266071 [2]青岛财经职业学校计算机系,山东青岛266100 [3]国网山东省电力公司昌邑市供电公司,山东昌邑261300

出　　处：《青岛大学学报（工程技术版）》2024年第1期87-94,共8页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：国家自然科学基金资助项目(52075278)。

摘　　要：针对现有的深度学习车辆目标检测算法检测效率和检测精度不高的问题,提出了基于YOLOv5模型的VOD-YOLOv5(Vehicle Object Detection-YOLOv5)的车辆目标检测算法。采用新型的轻量级卷积神经模块(c3f)提升网络的特征提取能力;引入Coord Conv代替原模型中的普通卷积模块增强对位置信息的感知能力;融入空间和通道融合的注意力机制Shuffle Attention,提升对模糊图像中小目标的检测精度。在UA-DETRAC车辆检测数据集上的实验结果表明,和原YOLOv5模型相比,VOD-YOLOv5模型的平均准确率均值(mean Average Precision,mAP)约提升了3.42%,对不同目标类检测的平均准确率(Average Precision,AP)均有提升,且检测速度满足实时性需求。Aiming at the problems of low detection efficiency and accuracy of existing deep learning vehicle object detection algorithms,based on the YOLOv5 model,the vehicle target detection algorithm of VODYOLOv5 was proposed.In this paper,a novel lightweight convolutional neural module(c3f)was adopted that could improve the feature extraction capability of the network;and introduced CoordConv instead of the ordinary convolution module in the original model,which enhanced the network’s ability to sense location information.The network model also incorporated Shuffle Attention,an attention mechanism for spatial and channel fusion,to improve the detection accuracy of the model for small targets in blurred images.The VOD-YOLOv5 model was validated on UA-DETRAC vehicle detection data set.The experimental results show that compared with the original YOLOv5 model,the mean average precision(mAP)of the VOD-YOLOv5 model proposed is increased by 3.42%,and the average precision(AP)of the detection of different target classes is improved,and the detection speed meets the real-time requirements,which effectively improves the detection performance of the vehicle object detection model.At the same time,it also proves the effectiveness and feasibility of the proposed algorithm.

关 键 词：深度学习 目标检测 YOLOv5 

分 类 号：TN822[电子电信—信息与通信工程]