基于YOLOv4的客车转向架部件漏油故障图像检测  

Image Detection of Oil Leakage Fault of Passenger Bus Bogie Parts Based on YOLOv4

作　　者：李海霞 徐彦恒 LI Haixia;XU Yanheng

机构地区：[1]兰州石化职业技术大学,兰州730060 [2]中国铁路兰州局集团有限公司兰州车辆段,兰州730000

出　　处：《科技创新与应用》2024年第8期49-53,共5页Technology Innovation and Application

基　　金：甘肃省高等学校创新基金项目(2022B-299)。

摘　　要：作为车体主要部件的铁路客车转向架是列车运行安全保障的关键部件,目前主要依靠客车故障轨旁图像检测系统检测出转向架故障并分类定位,但存在一定漏检和误检,检测准确率无法保证。基于此,针对客车转向架常见的关键部件漏油现象,展开漏油区域视觉图像缺陷检测研究,提出一种改进的YOLOv4目标检测算法,使用k-means++聚类方法获得更匹配关键部位漏油区域目标的候选框参数,更准确地识别和定位漏油区域目标。在网络中将部分标准卷积替换为可变形卷积,提高目标检测的准确性。As a main component of the carbody,the railway passenger car bogie plays a crucial role in ensuring the safety of train operations.Currently,the detection of bogie faults relies mainly on the trackside image monitoring system,which detects and classifies bogie faults.However,this system suffers from certain shortcomings,including missed detections and false alarms,leading to a lack of accuracy in fault detection.In response to this,this research focuses on the visual image detection of common oil leakage issues in key components of passenger car bogies.An improved YOLOv4 object detection algorithm was proposed,utilizing the k-means++clustering method to obtain candidate box parameters that better match the oil leakage areas in key components.This approach aims to achieve more accurate identification and localization of oil leakage areas.In the network,part of the standard convolution is replaced by deformable convolution to improve the accuracy of target detection.

关 键 词：客车转向架 深度学习 故障检测 YOLOv4 检测精度 

分 类 号：U279.323[机械工程—车辆工程]