基于机器视觉与YOLOv5的吊弦高清图像采集方法  

A High Identification Image Acquisition Method for Suspension String Based on Machine Vision and YOLOv5

作　　者：许建国 韩建民[1] 刘岩 王建超 XU Jianguo;HAN Jianmin;LIU Yan;WANG Jianchao(Beijing Jiaotong University,Beijing 100044,China;Dalian Weide Integrated Circuit Co.Ltd,Dalian,Liaoning 116085,China)

机构地区：[1]北京交通大学,北京100044 [2]大连维德集成电路有限公司,大连116085

出　　处：《铁道工程学报》2024年第9期81-86,共6页Journal of Railway Engineering Society

基　　金：中国中铁股份有限公司重大专项课题(2021-专项-07)。

摘　　要：研究目的:针对高铁接触网整体吊弦高清图像采集检测低效、准确率不高、投入成本高等问题,研究提出一种基于前端图像识别触发与深度学习算法融合定位的整体吊弦高清采集方法。研究结论:(1)采用嵌入式FPGA+ARM双处理平台,将基于YOLOv5的吊弦定位算法嵌入硬件结构中,通过FPGA实现图像运算硬件加速及识别过滤,通过ARM实现图像数据的高精度识别,整体吊弦图像检出率为99%,检时为4 ms/张,满足实时性检测要求;(2)通过FPGA与ARM的结合,内置专用图像处理算法芯片,滤除大量无用图像,获得清晰的整体吊弦图像,在数据集上训练和测试的准确率和召回率为100%,实际线路大于99%;(3)通过嵌入式FPGA+ARM硬件系统与YOLOv5的吊弦定位算法融合设计,减少了系统检测成本、扩大了技术应用范围,对推动接触网高质量智能建造与智能化运维具有借鉴作用。Research purposes:To address the issues of low efficiency,low accuracy,and high investment cost in high definition image acquisition and detection of the overall suspension string of high-speed rail overhead contact system,a high-definition acquisition method for the overall suspension string based on front-end image recognition triggering and deep learning algorithm fusion localization is proposed.Research conclusions:(1)Using an embedded FPGA+ARM dual processing platform,the YOLOv5 based suspension string positioning algorithm is embedded in the hardware structure.Image processing hardware acceleration and recognition filtering are achieved through FPGA,and high-precision recognition of image data is achieved through ARM.The overall suspension string image detection rate is 99%,with a detection time of 4 ms per image,meeting the requirements of real time detection.(2)By combining FPGA and ARM,a dedicated image processing algorithm chip is built-in to filter out a large number of useless images and obtain clear overall suspension string images.The accuracy and recall rate trained and tested on the dataset are 100%,and the actual circuit is greater than 99%.(3)By integrating the embedded FPGA+ARM hardware system with YOLOv5's suspension string positioning algorithm,the system detection cost has been reduced and the technical application scope has been expanded.This has a reference value for promoting high-quality intelligent construction and intelligent operation and maintenance of overhead contact system.

关 键 词：接触网 整体吊弦 机器视觉 深度学习 YOLOv5 

分 类 号：U226.51[交通运输工程—道路与铁道工程]