基于机器视觉的磁浮轨道功能件连接螺栓轻量化检测算法  

Lightweight Inspection Algorithm for Connecting Bolts of Maglev Guideway Functional Elements Based on Machine Vision

作　　者：彭菲 叶丰 黄诗华 PENG Fei;YE Feng;HUANG Shihua(National Maglev Transportation Engineering R&D Center,Tongji University,Shanghai 201804,China;State Key Laboratory of High-Speed Maglev Transportation Technology,Shanghai 201804,China)

机构地区：[1]国家磁浮交通工程技术研究中心,上海201804 [2]高速磁浮运载技术全国重点实验室,上海201804

出　　处：《交通与运输》2025年第1期31-37,共7页Traffic & Transportation

基　　金：国家重点研发计划项目(2023YFB4302502);上海市多网多模式轨道交通协同创新中心资助。

摘　　要：磁浮轨道功能件连接螺栓的表面锈蚀以及完备性检测是轨道运维工作的重要内容,但目前该项工作仍依赖于人工目测,效率低下且易漏检。为提高运维工作效率,针对完备性检测提出基于卷积神经网络算法的轻量化检测模型Bolts-net,并结合现场采集图像所制数据集验证模型的有效性。结果显示,采用Bolts-net的查准率为0.90、查全率为0.91、平均检测速率为97 FPS。同经典检测模型相比,在满足工程应用检测精度要求的同时,Bolts-net所包含的参数数目大幅减少,可部署运行于边缘设备,基于HSV色彩空间转换检测螺栓表面锈蚀区域,这对磁浮轨道运维具有重要的工程应用价值。The completeness and surface rust area detection of connecting bolts of maglev guideway functional elements is important for track operation and maintenance.It relies on manual visual inspection,which is inefficient and easily missed.To improve the efficiency of operation and maintenance work,based on the convolutional neural network algorithm,the lightweight detection model named Bolts-net is designed,combining with the dataset made from on-site collected images to check the model's validity.The results show that the precision of Bolts-net is 0.90,the recall is 0.91,and the average detection rate is 97 FPS.According to the requirements of detection accuracy for engineering application,the parameter number of Bolts-net is greatly reduced compared with the classical detection model,and can be deployed in the edge device,combined with the detection of surface corrosion areas on bolts based on HSV color space conversion,those work are of great value for the maglev track operation and maintenance in engineering application.

关 键 词：磁浮轨道运维 机器视觉 轻量化网络 注意力机制 螺栓检查 

分 类 号：U491[交通运输工程—交通运输规划与管理]