基于改进的YOLOv5安全帽佩戴检测算法  

Detection algorithm for wearing safety helmet based on improved YOLOv5

作　　者：雷建云[1,2,3] 李志兵 夏梦 田望[1,2,3] LEI Jianyun;LI Zhibing;XIA Meng;TIAN Wang(College of Computer Science,South-Central Minzu University,Wuhan 430074,China;Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprises,Wuhan 430074,China;Hubei Provincial Engineering Research Center of Agricultural Blockchain and Intelligent Management,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院,湖北武汉430074 [2]湖北省制造企业智能管理工程技术研究中心,湖北武汉430074 [3]农业区块链与智能管理湖北省工程研究中心,湖北武汉430074

出　　处：《湖北大学学报（自然科学版）》2024年第1期1-13,共13页Journal of Hubei University：Natural Science

基　　金：国家民委中青年英才培养计划(MZR20007);湖北省科技重大专项(2020AEA011);新疆维吾尔自治区区域协同创新专项(科技援疆计划)(2022E02035);武汉市科技计划应用基础前沿项目(2020020601012267)资助。

摘　　要：针对安全帽佩戴检测中存在的误检和漏检的问题,提出一种基于YOLOv5模型改进的安全帽佩戴检测算法。改进模型引入多尺度加权特征融合网络,即在YOLOv5的网络结构中增加一个浅层检测尺度,并引入特征权重进行加权融合,构成新的四尺检测结构,有效地提升图像浅层特征的提取及融合能力;在YOLOv5的Neck网络的BottleneckCSP结构中加入SENet模块,使模型更多地关注目标信息忽略背景信息;针对大分辨率的图像,添加图像切割层,避免多倍下采样造成的小目标特征信息大量丢失。对YOLOv5模型进行改进之后,通过自制的安全帽数据集进行训练检测,mAP和召回率分别达到97.06%、92.54%,与YOLOv5相比较分别提升了4.74%和4.31%。实验结果表明:改进的YOLOv5算法可有效提升安全帽佩戴的检测性能,能够准确识别施工人员的安全帽佩戴情况,从而大大降低施工现场的安全风险。Aiming at the problems of false detection and missing detection in helmet wearing detection,an improved helmet wearing detection algorithm based on YOLOv5 model was proposed.A multi-scale weighted feature fusion network was introduced in the improved model,which a shallow detection scale was added to the network structure of YOLOv5,and feature weights were introduced for weighted fusion to form a new four-foot detection structure,which effectively improved the extraction and fusion ability of shallow features in images.The SENet module was added into the BottleneckCSP structure of YOLOv5 Neck network,which made the model pay more attention to the target information and ignore background information.For large resolution images,image cutting layer was added to avoid the loss of large amount of small target feature information caused by multiple subsampling.After the YOLOv5 model being improved,the self-made safety hat data set was used for training and testing,and the mAP and recall rates reached 97.06%and 92.54%,respectively,which increased by 4.74%and 4.31%compared with YOLOv5.The experimental results show that the improved YOLOv5 algorithm can improve the detection performance of helmet wearing and accurately identify the wearing condition of the construction personnel,thus greatly reducing the safety risk on the construction site.

关 键 词：目标检测 多尺度加权特征融合 注意力机制 图像切割 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]