Research On Rail Surface Defect Detection Based On Improved YOLOv5

Due to the rapid growth of the total mileage and operating speed of China’s railways,as well as the fact that the environment in which the railways are located has almost experienced various typical global zones,terrains,and climates,achieving safe railway operation requires very high standards.Surface defects on railway tracks are usually caused by repeated movement of trains on the track,temperature changes,or natural disasters.If not properly maintained,the continuous growth of surface defects on the rail may lead to high maintenance costs and serious traffic accidents.At present,the surface defects of railway tracks mainly rely on manual inspection and railway track inspection vehicles.There are many manual operations and the work is tedious,and the quality and proficiency of the work vary from person to person.Therefore,studying emerging technologies that are conducive to automatic detection of surface cracks on railway tracks has important practical significance.This article studies the problem of rail surface defect detection based on deep learning.In response to the problems of poor universality,low accuracy,and low recall rate in rail surface defect detection methods,research and improvement are carried out on the basis of the original YOLOv5s,and the YOLOv5s CBE rail defect detection network is finally proposed.The main research content of this article is as follows:(1)This article provides an overview of deep learning based object detection algorithms,introduces their basic theory,structure,and classic models,and analyzes and studies CNN based classical object detection algorithms.It discovers the advantages,disadvantages,and existing problems of existing classical object detection algorithms.(2)We have self-made a dataset for detecting surface defects on railway tracks,including rail sleeper cracks and rail track defects.In response to the problems existing in its classic object detection algorithm,YOLOv5s was selected to detect surface defects on railway tracks.For YOLOv5s,its basic principle and network structure were introduced.(3)On the basis of YOLOv5s,research and improvement were carried out.Firstly,the CA attention module was added to the backbone C3 module and between C3 and SPPF,capturing channel relationships and position information from both channel and spatial dimensions,improving the feature extraction ability of YOLOv5s backbone network.Secondly,in the Neck section of YOLOv5s,a weighted bidirectional feature pyramid structure is used to fuse different scale information to obtain output feature maps with rich semantic information.At the same time,the bidirectional feature fusion pyramid structure optimizes the feature fusion effect by introducing weights to adjust the contribution of different scale input feature maps to the output.Finally,the Loss function CIoU in the original YOLOv5s is changed to EIoU.EIoU not only considers the center point distance and aspect ratio,but also considers the true difference between the width and height of the prediction box and the real box,which improves the prediction accuracy of the anchor box.Finally,the YOLOv5s CBE algorithm was proposed and experimental analysis was conducted on the two datasets proposed in this article.The proposed YOLOv5s CBE algorithm achieved detection accuracy of 80.7%on the rail sleeper crack dataset,and 76.3%on the rail track defect dataset,with a classification improvement of 3.7%and 7.7%.The improved algorithm model also brought slight lightness,reducing the false detection of defects,The issue of missed detections has been improved to varying degrees.