Research On Surface Defect Detection Of Railway Track Based On Image Processing

Over the past two decades,China’s railways have achieved leapfrog development.With the rapid growth of the total railway mileage,the continuous increase of high-speed passenger operation mileage and high-density shifts,the probability of damage to the rail surface has increased significantly.The surface damage of the rail will reduce the service life of the rail,and it will also easily induce the internal damage of the rail.Serious internal damage may even cause the rail to break,resulting in serious railway accidents such as train derailment.Therefore,there is a need for a railway inspection system that can quickly and accurately detect the surface damage of the rails,and repair them in time to avoid more serious damages,resulting in a significant reduction in the life of the rails.In this paper,the image processing technology is used to detect and identify the surface defects of the rails,and the key researches are carried out on the surface extraction of the rails,the feature extraction of the surface defects,and the target recognition.This paper mainly includes the following aspects.First of all,this paper studies and simulates three traditional image filtering methods.After comparing the denoising effects,the adaptive median filtering method with the best denoising effect is selected as the rail image denoising method;the gray distribution characteristics of the rail surface are analyzed,and a method for extracting rail surface with adaptive threshold is proposed.The effectiveness of this method is proved by simulation experiments.Secondly,the six kinds of rail surface defects are divided into two categories;carry out binary processing on the rail light strip,and according to the characteristics of the rail light strip,a preliminary judgment method for the edge defect of the light strip based on binarization is proposed;the light-band binary image is denoised and cracked closed by morphological operations.The edge defect features are extracted by the horizontal grayscale projection method and the waveform is smoothed by the moving average method.Simulation experiments show that this method can effectively extract the edge defect features.Then,according to the characteristics of block defects,an iterative threshold segmentation method based on image difference is proposed to segment block defects,and contour detection method is used to extract block defect features.Finally,select appropriate geometric features to distinguish defects,and construct a momentum BP neural network to realize the identification and classification of rail surface defects,Simulation results show that the neural network can accurately classify defects,and its network performance meets the detection requirements.Design defect detection software to implement the above method.