基于改进YOLOv5的火车连接钩舌识别方法  

Recognition method of train connection hook tongue based on improved YOLOv5

作　　者：王冬伟[1] 武彬 宋刚 高硕 韩昊宇 丁佳毅 董帆 万书亭 无 WANG Dongwei;WU bin;SONG Gang;GAO Shuo;HAN Haoyu;DING Jiayi;DONG Fan;WAN Shuting;无(Anhui Huadian Suzhou Power Generation Company Limited,Suzhou,Anhui 234000,China;Hebei Key Laboratory of Electric Machinery Health Maintenance&Failure Prevention,Baoding,Hebei 071003,China;Department of Mechanical Engineering,North China Electric Power University,Baoding,Hebei 071003,China)

机构地区：[1]安徽华电宿州发电有限公司,安徽宿州234000 [2]河北省电力机械装备健康维护与失效预防重点实验室,河北保定071003 [3]华北电力大学机械系,河北保定071003

出　　处：《河北工业科技》2024年第5期321-329,共9页Hebei Journal of Industrial Science and Technology

基　　金：国家自然科学基金(52275109);河北省自然科学基金(E2022502007)。

摘　　要：为了准确识别不同类型钩舌,确保自动复钩机器人能够根据火车连接钩舌状态实时调整机械臂的位姿,提出了一种基于改进YOLOv5的火车连接钩舌识别方法。首先,将YOLOv5主干网络中原有的C3模块替换为梯度流丰富的C2F模块(cross feature module),YOLOv5颈部网络中原有的C3模块替换为基于FasterNet模块构建的轻量化C3_FasterNet模块,并将CoordConv模块嵌入到YOLOv5的主干网络末端。其次,基于现场实测的火车连接钩舌图像进行了识别测试。结果表明:改进的YOLOv5算法在降低模型参数量的同时,可以有效提升对钩舌目标的检测精度,火车钩舌识别精度达到了98.7%,相较于原始算法,模型参数量减少了10.8%。研究结果为复钩机器人在执行钩舌复位和车厢连接操作方面提供了一种有效的解决方案。In order to accurately identify different types of train connection hook tongues and ensure that the automatic re-hook robot adjusts the pose of the robot arm in real time according to the status of the hook tongues,a train connection hook tongue recognition method based on an improved YOLOv5 model was proposed.First,the original C3 module in the YOLOv5 backbone network was replaced with the C2F module(cross feature module)of rich gradient flow,and the original C3 module in the YOLOv5 neck network was replaced with the lightweight C3_FasterNet module based on the FasterNet block,and the CoordConv module was embedded at the end of the YOLOv5 backbone network.Second,the recognition test was carried out based on the spot measured image of the train connection hook tongue.The results show that the improved YOLOv5 algorithm can effectively improve the detection accuracy of the hook tongue target while reducing the numbers of model parameters.The identification accuracy of the hook tongue reaches 98.7%,and the numbers of model parameters are reduced by 10.8%compared with the original algorithm,which can provide an effective solution for the re-hook robot in the operation of hook tongue resetting and carriage connection.

关 键 词：模式识别 图像处理 复钩机器人 火车钩舌 目标识别 YOLOv5 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]