基于自编码器的接触网定位器底座开口销缺失识别算法研究  

Research on recognition algorithm for missing split pins of the OCS locator base based on auto encoder

作　　者：于博轩 Yu Boxuan

机构地区：[1]中车工业研究院有限公司,北京100070

出　　处：《现代城市轨道交通》2021年第8期108-112,共5页Modern Urban Transit

摘　　要：随着接触网故障检测技术的发展,接触网悬挂状态检测系统可得到接触网定位器的海量高清图像用于线下故障排查。为提高接触网定位器底座开口销缺失检测的准确性和效率,基于自编码器无监督网络模型,根据定位装置开口销区域的图像特征设计开口销缺失识别算法,并通过实际现场图像数据验证算法的有效性和通用性。试验分析表明,该算法对于背景复杂、光照不均等现场实际图像均具有良好的适应性,算法检测准确率达90%以上,可提高接触悬挂零部件缺陷检测效率,具有重要的实际工程意义。With the development of OCS fault detection technology,the catenary suspension status detection system can obtain high-definition images of the catenary positioning device for offline troubleshooting.In order to improve the missing OCS locator base split pin detection,the split pin missing recognition algorithm based on auto encoder learning network model is devised to achieve this goal.This recognition algorithm detects special characteristics of the pin area on the OCS locator for the missing pin detection.The effectiveness and universality of the algorithm are verified by the actual image data.The experimental result analysis shows that the algorithm has a very important,practical engineering function.It adapts well to images with complex backgrounds and uneven illumination.The detection accuracy of the algorithm improves the defect detection efficiency of the suspension parts by more than 90%.

关 键 词：轨道交通 接触网检测 定位器 开口销缺失 自编码器AE NCC算法 支持向量机SVM 

分 类 号：U225[交通运输工程—道路与铁道工程]