Research On The Defect Recognition For Rail Surface Based On Image Processing

With the continuous development of railway transportation towards the direction of “high-speed passenger transport,heavy freight transport,and heavy truck density”,rails will be increasingly depleted,thus the number of defects on rail surfaces will also be added.Rail surface defects not only affect the normal driving of trains,but also may cause the defects develop to the rail internal injury,even worse,the rail may be cracked,and the security of railway transportation may also be threatened.Therefore,the timely and accurately detection of rail surface defects and effective maintenance mode can stifle potential risk factors in the bud.The traditional rail surface detection method is mainly through manual inspection,but the efficiency and accuracy of detection are not stable,the workload of inspectors is heavy and the working environment is dangerous.In this thesis,the image processing technology is used to detect the surface defects of rail,and the pre-judge of the defect of the rail surface.The segmentation of the target and the recognition of the target are mainly studied.The main research contents are as follows:Firstly,according to the requirements for the images of the rail surface detection,combining with the railway environment,selecting the appropriate camera,light source,computer,image acquisition software and other devices to build an experimental platform for the image acquisition of the rail surface.Based on the grayscale discrepancy between rail surface and non rail surface,rails surface extraction algorithm based on gray vertical projection is applied to extract the surface area of rails.The algorithm extracts rail surface area accurately and with high accuracy and fast speed.Secondly,through the qualitative analysis and quantitative analysis,several commonly used image enhancement algorithms are compared.Simulation results show that homomorphic filtering method has better effect on image enhancement.Then,after analyzing the enhanced image,it is found that the grayscale distribution between the adjacent images in the image sequence is analogous,and there is a grayscale discrepancy between the image defect area(or rail joint area)and the normal region.According to the statistical method,the pre-judgment algorithm of rail surface defect based on the difference of gray level is used to extract the suspected image containing defects.The simulation results show that this algorithm can extract the suspected image containing defects effectively.Thirdly,aiming at the problems that traditional C-V(Chan-Vese)model and LBF(Local Binary Fitting)model is sensitive to the initial contour position and has a slow convergence speed,a modified active contour model is built,which ameliorates non convex optimization model into convex model,and the Split Bregman iterative algorithm is used to solve the model of rail surface segmentation.This model,C-V model and LBF model are used to segment the rail surface swap blocks,cracks and rail joints respectively.The simulation results show that all the segmentation parameters of this model are better than those of the other two models,and it can be applied to the segmentation of the target of the rail surface.Finally,characteristics of the segmentation result are extracted.By means of statistics and observation,selecting the distinguish feature parameters of rail surface swap blocks,rail surface cracks and rail joints,and constructing the BP neural network to realize the recognition and classification of rail surface swap blocks,rail surface cracks and rail joint.The simulation results show that the recognition network is accurate in classification and can meet the requirements of detection.