Research On Recognition Of Rail Fastener Rust Based On Color Image Multi-threshold Segmentation Algorithm

The rail fasteners rust defect on surface is one of the important reasons for railway traffic accidents.The detection and maintenance of rusted rail fasteners can greatly reduce the occurrence of railway traffic accidents.The rapid development of computer image processing technology provides a new technical processing method for the rapid and accurate detection of rust defects on the surface of rail fasteners.However,the surface rust of fasteners is characterized by random distribution,different sizes and discontinuous distribution.Meanwhile,detection would suffer from uneven illumination,shadow as well as other background noises,which makes it difficult to accurately detect fastener rust defects.In order to overcome these difficulties and improve fastener rust defect detection accuracy,this thesis proposes a color image adaptive multi-threshold segmentation combined with image features of rail fastener rust defect detection algorithm.The specific contents are as follows.In the stage of image preprocessing,an Improved Pyramid Evolution Strategy(IPES)is proposed.This thesis designs an adaptive search operator suitable for the multi-threshold segmentation problem of color images,expands the search space at all levels,improves the optimization ability of the algorithm;uses competition and cooperation relationship between populations to solve the local optimization problem.At the same time,it takes Otsu as the optimization goal function to sharpen image characteristics and improve image quality.The existing standard test images are used to test algorithm performance and compared with the other eight algorithms.The experimental results show that peak signal-to-noise ratio and structural similarity of the segmented image by IPES algorithm are better than those of the contrast algorithm,indicating that IPES algorithm has good performance in image segmentation and can improve the image quality.In the stage of image defect detection,two weak classifiers are constructed based on color and texture features.The first classifier is roughness analysis,calculating the gray level co-occurrence matrix of image,setting the roughness threshold by statistical analysis and comparing with uniformity value of the candidate region to detect whether image has defects.The second classifier is rust color analysis.Firstly,rust color spectrum is constructed by a large number of rust images,and center and standard deviation of rust color domain are calculated by rust color spectrum.Then,the pixels of rust defect image are mapped in RGB color space,and the deviation between the pixels in RGB color space and the center of rust color domain is calculated.The size of deviation and standard deviation can be judged to determine whether the image contains rust defects or other defect types.Finally,the accuracy and recall rate are defined to evaluate classification performance of classifier.The experimental results show that two evaluation indexes have better results,which verifies effectiveness and accuracy of classifier in defect detection.This thesis proposes rail fasteners rust defect detection algorithm,which uses color image adaptive multi-threshold segmentation algorithm to sharpen image characteristics and improve image quality in the image preprocessing stage,and constructs a color and texture-based classifier to improve images rust detection accuracy in the defect detection stage.This algorithm not only provides a new idea for the field of image segmentation,but also provides a new method for solving railway fasteners corrosion identification problem.At present,this algorithm has been applied to railway fasteners rust detection and has broad application prospects.