基于MFDC-SSD网络的接触网定位线夹缺陷识别  

Fault Identification of Overhead Contact System Steady Ears Based on MFDC-SSD Network

作　　者：屈志坚[1,2] 张博语 杨行 李迪 QU Zhijian;ZHANG Boyu;YANG Hang;LI Di(State Key Laboratory of Performance Monitoring and Protecting of Rail Transit Infrastructure,East China Jiaotong University,Nanchang 330013,China;School of Eletrical and Automation Engineering,East China Jiaotong University,Nanchang 330013,China)

机构地区：[1]华东交通大学轨道交通基础设施性能监测与保障国家重点实验室,江西南昌330013 [2]华东交通大学电气与自动化工程学院,江西南昌330013

出　　处：《铁道学报》2024年第5期48-57,共10页Journal of the China Railway Society

基　　金：江西省自然科学基金(20232ACB204025);江西省高层次高技能领军人才培养工程(202223323);轨道交通基础设施性能监测与保障国家重点实验室自主课题(HJGZ2022203)。

摘　　要：针对接触网应用环境复杂,定位线夹体积小、安装方向特殊、不易识别,传统的目标检测算法效果较差等问题,基于多尺度特征融合密集连接网络模型,提出了一种接触网定位线夹缺陷检测新方法。首先结合DenseNet与Inception模块对SSD模型的特征提取网络进行改进,在特定的特征层间共享上下文信息,提升特征提取能力。然后从网络深层到浅层,逐级采用FPN融合SSD检测的多尺度特征图,设计多特征融合的密集连接网络模型,最后将f GIoU作为边框损失函数,在训练中优化真实框和预测框的重合度。对采集的某段接触网定位线夹图像数据集进行检测识别。结果表明:该定位线夹缺陷检测方法可在复杂接触网背景下,对定位线夹脱落和松动进行检测,且在不同角度、亮度的图像中均具有较强的鲁棒性。Aiming at the problems of complex application environment of overhead contact system,small size of steady ears,special installation direction,difficult identification,and poor performance of traditional target detection algorithms,a new method for detecting defects in steady ears of overhead contact system was proposed based on multi-scale feature fusion dense connectivity network(SSD)model.Firstly,the feature extraction network of SSD model was improved by combining DenseNet and Inception modules to share the context information among specific feature layers to enhance the feature extraction capability.Then the multi-scale feature maps for SSD detection were fused using FPN step by step from deep to shallow layers of the network to design a dense connected network model for multi-feature fusion.Finally,f GIoU was used as the edge loss function to optimize the overlap between the real and predicted frames during training.The collected image dataset of steady ears of a section of overhead contact system was detected and recognized.The results show that the steady ear defect detection method designed in this paper can detect the steady ear detachment and loosening under the complex backgrounds of the overhead contact system with strong robustness in images with different angles and brightness.

关 键 词：接触网 缺陷识别 定位线夹 特征金字塔 边界框损失函数 

分 类 号：TM73[电气工程—电力系统及自动化] U225[交通运输工程—道路与铁道工程]