城轨交通接触网悬挂状态智能检测算法及应用  

Intelligent Detection Algorithm and Application of OCS Suspension Status in Urban Rail Transit

作　　者：陈茹 CHEN Ru(Technology Department,Chengdu Branch of Beijing Dinghan Testing Technology Co.,Ltd.,Chengdu Sichuan 610000,China)

机构地区：[1]北京鼎汉检测技术有限公司成都分公司技术部,四川成都610000

出　　处：《中国铁路》2024年第2期161-167,共7页China Railway

摘　　要：通过高清成像对接触网悬挂状态进行检测,及时发现零部件松脱卡磨断等缺陷问题并维修,是保障城轨交通接触网系统安全运行的重要手段。采用YOLOV4的最小版本YOLOV4-tiny作为检测算法的基础模型,将原有Dropout层去掉,将上层特征直接经过卷积层后输入到下层卷积层,并加入Mosaic数据增强方式提高算法的泛化性,同时将原有Leaky-Relu函数替换为更加平滑的Mish函数。接触网缺陷定位的流程为:根据改进后的YOLOv4-tiny算法训练得到零部件区域定位模型,用于定位零部件的位置;再根据改进后的YOLOv4-tiny算法训练得到缺陷识别模型,用于判断该零部件是否有缺陷,并给出缺陷的具体位置和缺陷类别。实际应用表明:该算法能准确、快速定位零部件缺陷位置,并给出缺陷类型,总精度达90%以上,大大地降低了检测耗时与成本,并保障了作业人员安全。It is an important means to ensure the safe operation of OCS in urban rail transit by detecting the suspension state of OCS through high-definition imaging,and timely finding out defects such as looseness,jamming and wear of components and repairing them.The smallest version of YOLOV4-tiny is used as the basic model of the detection algorithm.The original Dropout layer is removed,and the upper-layer features are directly input to the lower convolutional layer after passing through the convolutional layer.Mosaic data enhancement method is added to improve the generalization of the algorithm,and the original Leaky-Relu function is replaced with a smoother Mish function.The process of OCS defect locating is as follows:According to the improved YOLOv4-tiny algorithm training,a component area locating model is obtained for locating the position of components;then according to the improved YOLOv4-tiny algorithm training,a defect recognition model is obtained for judging whether the component has defects and giving the specific location and category of defects.Practical application shows that the algorithm can accurately and quickly locate the defect position of components,and give the type of defects with a total accuracy of above 90%.It greatly reduces the detection time and costs and ensures the safety of operation personnel.

关 键 词：城轨交通 接触网 悬挂状态 智能检测 YOLOV4-tiny 深度学习 缺陷检测 

分 类 号：U225[交通运输工程—道路与铁道工程] TP391[自动化与计算机技术—计算机应用技术]