Research On Defect Recognition Of Rail Fastener Based On Image Processing

In recent years,the railway operation has been developing towards the direction of high speed and overloading.It put higher demands on the maintenance of rail-tracks and the security detection.As one of the main components in the railway infrastructures,damage to the railway track fasteners or lost fasteners can lead to major accidents such as train derailment.As an indispensable item in the railway monitoring system,the defect detection of fasteners is very important.In order to change the current situation of the use of manual visual inspection along the railway in China,it is particularly important to develop a non-contact automatic detection of railway fasteners.In view of the shortcomings of the existing railway fastener detection algorithm,the thesis takes advantage of image processing to detect defects in railway fasteners.The main research contents of the thesis are as follows:First,collecting railway fastener images and filtering them.According to the experimental requirements,selecting the light source,camera,and lens that meet the research needs of the thesis and collect fastener image.The image is filtered using a median filter whose template is 3×3.The simulation experiments show that the filtering method has a good filtering effect and can well protect the defect details of the fastener image.Second,the edge detection and positioning of railway fasteners.In view of the shortcomings of the traditional Canny algorithm,a method of edge detection based on improved Canny algorithm is proposed in this thesis.When calculating the gradient,adding the gradient magnitude of the 45 degrees direction and the 135 degrees direction,effectively avoid missing some pixels.Simultaneously,introducing the two-dimensional Otsu threshold method into the algorithm,enables the algorithm to adaptively select the optimal threshold based on the unique characteristics of the fastener itself.This method effectively avoids intermittent edges and false edges.Therefore,the purpose of improving the edge quality of the detection and perfecting the edge information is achieved.After extracting the edge information,the Hough transform is used to propose a straight line method to locate the track fastener region.The simulation experiments show that improved algorithm can extract more complete edge information and increase the accuracy of positioning.Finally,extracting the PHOG features and MBLBP features of the railway fastener images.After combining these two features,input SVM for classification and recognition.Compared to the limitations of a single feature,fusion of MBLBP features for extracting texture information and PHOG features for extracting spatial edge information,the new features have combined the advantages with both.The simulation experiments show that compared with a single feature,the accuracy obtained by using the fused feature as aclassification basis is significantly improved.