基于Swin Transformer的YOLOv5安全帽佩戴检测方法  

YOLOv5 Helmet Wearing Detection Method Based on Swin Transformer

作　　者：郑楚伟 林辉[1] ZHENG Chuwei;LIN Hui(College of Intelligent Engineering,Shaoguan University,Shaoguan 512005,China)

机构地区：[1]韶关学院智能工程学院,广东韶关512005

出　　处：《计算机测量与控制》2023年第3期15-21,共7页Computer Measurement &Control

基　　金：广东大学生科技创新培育专项资金资助项目(pdjh2022b0470)。

摘　　要：针对目前施工现场的安全帽检测方法存在遮挡目标检测难度大、误检漏检率高的问题,提出一种改进YOLOv5的安全帽检测方法;首先,使用K-means++聚类算法重新设计匹配安全帽数据集的先验锚框尺寸;其次,使用Swin Transformer作为YOLOv5的骨干网络来提取特征,基于可移位窗口的Multi-head自注意力机制能建模不同空间位置特征之间的依赖关系,有效地捕获全局上下文信息,具有更好的特征提取能力;再次,提出C3-Ghost模块,基于Ghost Bottleneck对YOLOv5的C3模块进行改进,旨在通过低成本的操作生成更多有价值的冗余特征图,有效减少模型参数和计算复杂度;最后,基于双向特征金字塔网络跨尺度特征融合的结构优势提出新型跨尺度特征融合模块,更好地适应不同尺度的目标检测任务;实验结果表明,与原始YOLOv5相比,改进的YOLOv5在安全帽检测任务上的mAP@.5:.95指标提升了2.3%,检测速度达到每秒35.2帧,满足复杂施工场景下安全帽佩戴检测的准确率和实时性要求。Aiming at the problems of difficult detection of occluded objects, high false detection and missed detection rate in current helmet detection methods on construction sites, an improved YOLOv5 helmet detection method is proposed in this paper. Firstly, a K-means++ clustering algorithm is used to redesign the prior anchor box size to match the helmet dataset. Secondly, Swin Transformer is used as the backbone network of YOLOv5 to extract features. The multi-head self-attention mechanism based on shiftable windows can model the dependencies between different spatial location features, effectively capture the global context information, and have better the feature extraction capability. Thirdly, a C3-Ghost module is proposed to improve the C3 module of YOLOv5 based on Ghost Bottleneck, and generate more valuable redundant feature maps through low-cost operations, which effectively reduces model parameters and computational complexity. Fourthly, a new feature fusion module is proposed based on the structural advantages of cross-scale feature fusion in bidirectional feature pyramid network, which can better adapt to the target detection tasks of different scales. The experimental results show that compared with the original YOLOv5, the mAP@.5:.95 index of the improved YOLOv5 on the helmet detection task is improved by 2.3%, and the detection speed reaches 35.2 frames per second, which meets the accuracy and real-time requirement of the helmet wearing detection in complex construction scenarios.

关 键 词：安全帽佩戴检测 YOLOv5 Swin Transformer GHOST 新型跨尺度特征融合 K-means++ 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]