Research On Intelligent Identification Method Of CRTS Ⅱ Rail Slab Cracks Based On Machine Vision

With the network development of China’s high-speed railroads and the increasing mileage,personnel safety issues and maintenance costs have become critical issues that need to be solved.Among them,the timely detection and maintenance of high-speed railroad ballastless track injury and damage has become an important issue for high-speed railroad maintenance and repair department,and its accuracy and timeliness are related to the operational safety of high-speed railroad.At the present stage,China mainly adopts manual inspection for CRTS Ⅱ ballastless track slab injury and damage detection,which leads to low efficiency,high cost and insufficient safety due to insufficient light at night,harsh working environment and subjective judgment of detection results.The use of machine vision technology for high-speed railroad ballastless track slab crack damage detection can greatly improve the accuracy and efficiency of the detection work.This paper proposes a detection method for cracks on the surface of CRTS Ⅱ ballastless track slab of high-speed railroad,and quantifies and analyzes the detection results to achieve intelligent,high-efficiency and high-precision detection.The main research contents are as follows:1)Describe the background and significance of this paper,analyze and summarize the research on cracks at home and abroad,list the advantages and disadvantages of different detection methods,and analyze the feasibility of using machine vision for CRTS Ⅱ ballastless track slab crack detection.To summarize the types of cracks in CRTS Ⅱ ballastless track slab and their distribution locations,design the data collection platform,build the test collection platform,design the data collection scheme,build the track slab crack data set and splice the left and right rails for the collected data.2)In order to locate the cracked rail slab efficiently and accurately,a method based on improved Faster R-CNN is proposed to detect the cracks of CRTS Ⅱ ballastless track slabs.In order to reduce the redundant anchor frames and improve the detection target,the Soft-NMS algorithm is used to improve the crack detection effect.The evaluation criteria of CRTS Ⅱ ballastless track slab cracks are established,and the improved method is compared with R-FCN,YOLOv5,Faster R-CNN and YOLOX network based on the evaluation criteria.3)In order to extract the crack contour,the improved DeepLabV3+ based rail slab crack segmentation method is proposed.The method performs full-pixel segmentation of the detected cracks based on the crack detection results,increases the degree of scale fusion and reduces the number of downsampling in order to solve the segmentation edge blurring problem,and uses hybrid cavity convolution in order to improve the tessellation lattice effect generated by the excessive expansion rate of cavity convolution.The specific pixel information of the cracks is extracted using the improved method in order to evaluate their morphology in the subsequent work.The method is compared with four other segmentation methods,U-net,Seg-net,Deepcrack,and DeeplabV3+.It is verified that the proposed method performs well on the dataset.4)For quantitative analysis of cracks,based on the crack segmentation,the target skeleton line is extracted and the crack length is calculated,and the crack width is calculated using the orthogonal skeleton line method.The specific implementation process includes the following steps: firstly,the original image is pre-processed,the connected domain of the binary image is labeled and analyzed,and the area where the total number of pixels in the connected domain is less than the specified threshold is eliminated to retain the real crack information and remove the interference noise;secondly,the Marching Squares algorithm is used to find the contours in the image,and then the median transformation is performed to obtain the single-pixel Finally,the normal vector is used to predict the skeleton line and calculate the intersection of the normal and crack contour to obtain the accurate crack width data.The method is rigorous and scientific,which can ensure the accuracy and reliability of the calculation results.In this paper,we take the collection of railroad field track slab pictures as the starting point,and carry out the tests of target detection,semantic segmentation and quantitative analysis on them in turn.Targeted optimization is proposed for the target of CRTS Ⅱballastless track slab cracks.The improved network is compared with the mainstream network,and the advantages of the research method in this paper in terms of accuracy rate are proved.To fill the gap of quantitative analysis for CRTS Type II ballastless track slab crack detection,a set of processes is proposed for reference.