基于深度学习的雷达交通目标检测研究  

Research on radar traffic target detection based on deep learning

作　　者：汪赟杰 谭爱红 WANG Yunjie;TAN Aihong(School of Mechanical and Electrical Engineering,China Jiliang University,Hangzhou 310000,China)

机构地区：[1]中国计量大学机电工程学院,浙江杭州310000

出　　处：《现代电子技术》2023年第21期134-140,共7页Modern Electronics Technique

基　　金：浙江省自然科学基金资助项目(Y21F010057)。

摘　　要：针对将深度学习应用于交通场景下的雷达距离多普勒谱图目标检测任务时,交通目标尺寸小、特征不明显导致目标检测算法出现漏检、误检的问题,提出一种改进的YOLOv5⁃KFCS模型。首先提出基于K⁃means++聚类Anchor生成方法,确定最优Anchor尺寸,实现Anchor与实际目标的精准匹配;然后在模型中添加改进的FCBAM注意力模块,增强模型对于模糊目标和小尺寸目标特征的提取能力;接着将CARAFE作为上采样模块,提升网络对背景噪声的过滤能力以增强小目标特征的表征能力;最后将Swin Transformer模块引入到网络末端C3模块中,改善模型网络末端特征图分辨率低的问题。实验结果表明,改进后的YOLOv5⁃KFCS有效改善了漏检、误检问题,相较基准YOLOv5s平均检测精度提高5.3%,达到了93.5%,检测速度为70 FPS,满足检测实时性,并且综合性能优于其他方法。When deep learning is applied to the task of radar distance Doppler spectrogram target detection in traffic scenes,the small size and inconspicuous features of traffic targets lead to the missed and false detection in the target detection algorithm,so an improved YOLOv5⁃KFCS model is proposed in this paper.An anchor generation method based on K⁃means++clustering is proposed to determine the optimal anchor size and achieve accurate matching between anchor and actual targets.An improved FCBAM attention module is added to the model to enhance the extraction capability of the model for fuzzy targets and small⁃size target features.CARAFE is used as an up⁃sampling module to improve the network's filtering ability to background noise,so as to enhance the characterization ability of small target features.Finally,the Swin Transformer module is introduced into the C3 module at the end of the network to increase the resolution of the feature map at the end of the model network.The experimental results show that the improved YOLOv5⁃KFCS can eliminate the missed and false detection effectively,its average detection accuracy(reaching 93.5%)is improved by 5.3%in comparison with the benchmark YOLOv5s,and its detection speed is 70 FPS,which satisfy the detection real⁃time performance,and its comprehensive performance is better than the other methods.

关 键 词：深度学习 雷达 目标检测 距离多普勒谱图 YOLOv5s 交通场景 

分 类 号：TN911.73-34[电子电信—通信与信息系统]