Research Of The Detection Algorithm On Railway Fastener Defects Based On SVM

With the continuous development of transportation in China,railway transport has become an important part of the development of the national economy.This mode of transportation,which has covered all over the country,has greatly promoted the development of cities along the railway line.It has made great contribution to the economic construction of our country.At the same time,the safety problem of railway transportation is inevitably exposed in the presence of technicians.Railway lines are mainly composed of sleepers,rails and fasteners.The fastener is the key component connecting the sleeper and rail,which is used to prevent the longitudinal and lateral movement of the rail and ensure the safe and smooth operation of the train.Therefore,intact fasteners play a very important role in the normal operation of rail and sleeper,while damaged or missing fasteners will bring great trouble to train safety.At present,the inspection of railway fasteners mainly depends on manpower,that is to say,the integrity of the fasteners can be judged by the visual observation of the patrol personnel and the work experience.This method has great dependence on the natural environment,the attention and proficiency of the inspectors,and because it can not ensure the personal safety of the staff,it is easy to affect the on-time operation of the train,and so on.In the booming environment of railway transportation,it can not meet the requirements of modern railway line detection.Therefore,it is urgent to develop an automatic detection method for railway fasteners.Based on the literature about computer vision technology and railway fastener inspection,this dissertation aims at the key features of railway fastener in our country,and extracts the localization of fastener image by BEMD.Combined with the advantages of efficient and easy of SVM,a method of railway fastener defect detection based on SVM is proposed.The main contents of this dissertation are as follows:The images captured by the high-speed camera usually contain multiple targets,so it is necessary to distinguish the fasteners in different positions.The localization model of fastener image is constructed by BEMD and Hough transform.In this model,the obtained RGB color fastener image should be pre-processed operations,so as to prepare for BEMD.Then the gray image of railway fastener is decomposed into four IMFs and one residual by BEMD.After reconstruction of IMF1 image and IMF2 image,the fusion of detail information and illumination information of fastener can be obtained.The edge image of railway fastener can be extracted by binarization of reconstructed image and morphological operation such as thinning and removal.Based on the previous localization method of fastener region,it is creatively proposed to lock the position of railway fastener in the form of circle according to the curve feature of fastener image and Hough transform.According to the relationship with rail position and the size of railway fastener,it is further modified.By cutting out the specific location of the lock and finally extracting the individual fastener image and its edge image from the image,it is convenient for the subsequent feature extraction and classification.We compare the performance of advanced PHOG feature and LOOP feature in fastener extraction in the field of image recognition,combine their respective advantages to achieve loose coupling,and get the PHOG-LOOP fusion feature.Among them,PHOG feature is a multi-level feature developed on the basis of HOG feature,while LOOP feature is the advantage fusion of LBP operator and LDP operator,which is based on LBP and Kirsch template of LDP to extract image features.The fusion of them not only obtains the structural features of the fasteners,but also obtains the texture features of the fasteners.At last,the key features used to distinguish are determined by reducing the dimension by PCA.The key innovation of this dissertation is to solve the parameter selection optimization problem of SVM.In algorithm,large parameter selection will lead to poor learning ability;small parameter selection will bring about the problem of "over-learning",so parameter selection is very important to the performance of SVM.In order to improve the ability of classification,the optimization of parameter groups is realized by improved particle swarm optimization algorithm.On the one hand,the learning factor of comparing with the neighborhood in the search process is increased,and the factors that influence each other in the search process are considered in the optimization process.On the other hand,the optimal value is quickly induced by adaptive inertia weight.The simulation results show that the speed and efficiency of the optimization algorithm are fully verified by the representative data set provided in UCI.A classification model of railway fastener defect detection is designed based on SVM.From the pre-processing and localization of railway fastener image to the optimization of classification algorithm,and then to the final recognition and classification,a complete detection flow is realized.Based on the proposed model,the recognition and classification experiments of intact,damaged and lost railway fastener images are carried out by using MATLAB on the Windows platform.The experimental results show that the defect detection of railway fastener based on SVM can realize the intelligent identification of the working state of railway fastener,which provides a new method for the intelligent development of railway construction.