Research On Application Of Image Recognition Technology In Anomaly Detection Of Rail Fasteners

Railway transportation has always played an important part in people's lives.From travel to the delivery of goods,rail transport has become an indispensable role in daily life.The track is not only the basic equipment for the railway transportation but also an important part of the railway line.Its performance related to the comfort and safety of the traffic directly,and it relates to the service life of the railway line and the rolling stocks.Due to the fact track equipment is exposed to various natural environments all year round gone through the testing by the weather and natural environment,also because of the repeated use of locomotives,the rail fasteners and the track will be deformed and damaged,which requiring inspection and maintenance promptly.It is a great significance to maintain the stability and security in the process of railway transportation.Among them,one of the focuses of railway safety is the inspection of railway infrastructure,and the anomaly detection of rail fasteners is also very important.With the rapid development of computer technology and the improvement of image processing technology,computer vision technology has been introduced into railway detection systems and has become a real-time and highly efficient detection method.This paper proposes an image recognition technology based on the classification of rail fasteners and identifies the abnormal state of the fasteners.Perform image acquisition of rail fasteners at first,and then pre-processed the collected images,including the image greying and enhancement;In image filtering,since there are many interferences in the image of rail fasteners,especially the salt-and-pepper noise,an improved adaptive filtering method is used to process the image;secondly,the position of the pillow shoulder is determined by edge detection to determine the rail buckle.Segment the fasteners in the image to get the image of the target area.The HOG features of the fastener images are extracted,and experiments are performed by using a linear SVM as a classifier.Because the redundant information in the HOG feature extraction has too much influence on the recognition rate and classification speed,this paper designs a method based on dimension reduction and feature composite to improve the recognition rate and classification speed.Firstly,perform dimensionality reduction on HOG feature based on the principal component analysis.Secondly,we will fuse the HOG features and LBP features after the dimension reduction.The feature of LBP is used to extract the texture information from the image.This feature is complementary to the HOG features.The extracted LBP features and HOG features are combined in a cascade manner.Finally,linear SVM is used for classification.In this paper,the characteristics of the appeal are extracted and tested,and the data obtained under different special extractions are obtained,including the recognition rate and classification speed of the HOG feature,LBP feature,PCA-HOG feature,and PCA-HOG+LBP feature in each case.After comparison,it is concluded that the best performance is obtained under PCA-HOG+LBP feature extraction,and the feasibility of the design scheme of this article in the detection of rail fasteners is also verified.