基于改进YOLOv5s算法的钢轨扣件状态检测方法  

Rail Fastener Status Detection Method Based on Improved YOLOv5s Algorithm

作　　者：陈修忻 CHEN Xiuxin(Track Works Branch of Shanghai Metro Maintenance Support Co.,Ltd.,200233,Shanghai,China)

机构地区：[1]上海地铁维护保障有限公司工务分公司,上海200233

出　　处：《城市轨道交通研究》2024年第S01期147-151,共5页Urban Mass Transit

摘　　要：[目的]为提高轨道巡检效率以及优化巡检策略,有效避免漏巡、漏检等问题的发生,提出了一种基于改进YOLOv5s算法的钢轨扣件状态检测方法。[方法]介绍了YOLOv5算法的网络结构。在YOLOv5s算法的骨干网络中融入C3-CBAM(卷积注意力)模块以获取更多细节特征,然后采用BiFPN(加权双向特征金字塔)网络进行多尺度特征融合,形成改进YOLOv5算法。针对弹条断裂、弹条缺失、弹条移位和螺栓缺失4种状态进行了试验验证。[结果及结论]采用改进后的YOLOv5s算法比原YOLOv5s算法在测试精度、召回率、平均精度上都有所提高,表明该方法对钢轨故障扣件分类检测具有很好的工程应用价值。[Objective]To enhance the efficiency of track inspection,optimize inspection strategies,and effectively prevent issues such as missed inspections and detections,a method based on the improved YOLOv5s algorithm for detecting the status of rail fasteners is proposed.[Method]The network structure of the YOLOv5 algorithm is presented.The C3-CBAM(convolutional attention)module is incorporated into the backbone network of YOLOv5s algorithm to capture more detailed features.Subsequently,a BiFPN(bi-directional feature pyramid network)network is employed for multi-scale feature fusion,forming the improved YOLOv5 algorithm.Experimental validation is conducted for four states:rail clip fractures,rail clip loss,rail clip displacement,and bolt loss.[Result&Conclusion]The improved YOLOv5s algorithm shows improvements over the original YOLOv5s algorithm in terms of test accuracy,recall rate,and average precision,indicating its significant engineering application value for rail fastener fault classification detection.

关 键 词：城市轨道交通 轨道 钢轨扣件 分类检测 YOLOv5s算法 注意力机制 加权双向特征金字塔 

分 类 号：U213.53[交通运输工程—道路与铁道工程]