Research On Key Technologies Of Image Segmentation Of Rail Surface Defects

The rapid development of the modern railway industry is faced with the difficulties and challenges of the ever-increasing running speed,operating mileage,and heavy load of trains,which increases the potential safety hazards of rail infrastructure.Rails are an important part of track infrastructure.Due to high-density operation,high-load loads and the external natural environment,certain abrasion damage will occur on the surface,which affects the comfort and safety of trains.Timely detection of track surface defects,providing reliable and real data to relevant departments and units,and real-time detection and maintenance of the health status of track surface defects to ensure the reliability,safety and service life of the rails,and for the normal operation of the rails and maintenance are of great practical significance.In this thesis,some research on the key technologies of image segmentation of rail surface defects is carried out,and the main contributions are as follows:In the preprocessing of the track surface image,bilateral filtering is used to denoise the track surface image,which improves the denoising performance of the track surface image,and better protects the edge detail information of the image,which is used for the extraction of the track surface area and the segmentation of the defect surface.,detection,identification and classification and other follow-up work has laid a good foundation and has certain practical value.A method for extracting track area based on S-component of HSV color space is proposed.The actually collected track images include not only the track surface area,but also interference areas such as sleepers,fasteners,and gravel.In order to reduce the difficulty of track surface defect detection and shorten the track surface defect detection time,it is necessary to accurately and quickly extract the track surface area.The traditional rail surface area extraction methods need to pre-determine the width of the rail surface to varying degrees,and are easily disturbed by uneven illumination.In the RGB space image,the regularity of the gray value distribution of the track image is not strong,and it is difficult to overcome the interference such as uneven illumination.In order to solve the above problems,the S-component image in HSV space image has the characteristics of insensitivity to illumination interference and strong resistance to uneven illumination.First,the RGB track image is converted into the HSV color space track image,and the S-space component image is extracted separately.,in order to overcome the interference caused by the changes of illumination conditions and noise on the extraction of the rail surface area,and at the same time,the improved linear function is used to enhance the Sspace component image,effectively enhancing the difference between different areas;finally,by drawing gray Degree projection curve,search for the maximum value and the second maximum value on the left and right sides of the column curve,determine the left and right boundaries of the rail surface area,and complete the extraction of the rail surface area.An improved PSO 2D-Otsu track surface defect segmentation algorithm is proposed.In view of the shortcomings of the two-dimensional maximum inter-class variance threshold method(2D-Otsu),such as slow convergence speed,low optimization accuracy and easy to fall into local optimum,the reciprocal value of the two-dimensional Otsu method’s inter-class variance is set as the fitness function of the particle.,using chaotic map to enhance the randomness and uniformity of the initial distribution of particle population,introducing the average optimal fitness value and adaptively adjusting the inertia weight to update the position of individual particles,avoiding premature convergence of the algorithm.The mutation rate is constructed,combined with the Cauchy mutation operator,to increase the diversity of the population and help the algorithm to jump out of the local optimum.Simulation experiments show that the algorithm in this thesis has better global optimization ability,convergence and optimization accuracy,and the effect of segmenting track surface defect images is ideal.