基于4C图像弱监督检测的接触网开口销检测方法  

作　　者：宋东海 王志鹏 覃雅兰 缪弼东 

机构地区：[1]中铁电气化局集团有限公司京沪高铁维管公司 [2]北京交通大学先进轨道交通自主运行全国重点实验室

出　　处：《电气化铁道》2023年第S02期23-27,共5页Electric Railway

摘　　要：接触网悬挂状态检测监测装置(4C检测装置)的应用有效指导了接触网设备的维修,提高了铁路行车安全性。除了传统的人工检测,目前大量的开口销图像依靠全监督视觉方法检测,且开口销在图像中占比较小,因此数据标注过程所需人力成本较大。本文提出基于弱监督的4C图像开口销检测方法,旨在仅对训练数据进行图像级标注即可完成目标检测任务,达到完善接触网开口销检测手段、提升检测效率及智能水平的目的。The application of inspection and monitoring device for OCS suspension status(4C inspection device)effectively guides the maintenance of OCS equipment and improves the safety of railway operation.In addition to traditional manual inspection,a large number of split pin images currently rely on fully supervised visual methods for inspection,and the proportion of split pins in the images is relatively small.Therefore,the labor cost required for the data annotation is relatively high.The paper puts forward a 4C image split pin inspection method based on weak supervision,aiming to complete the target inspection task by only annotating the training data at the image level,achieving the goal of improving the inspection methods of OCS split pins,improving inspection efficiency and intelligence level.

关 键 词：弱监督检测 深度学习 4C检测装置 开口销检测 

分 类 号：U226.8[交通运输工程—道路与铁道工程]