Research On Surface Defect Classification Of Contact Track Video

At present,the service status and defect detection mode of the contact rail in China is still mainly manual.The inspectors need to walk along the rail line with professional contact rail measuring ruler(three rail ruler),reflective visual equipment and auxiliary lighting fixtures,and conduct relevant measurement while walking.Because the detection distance is generally long,and the detection tool does not have the automatic detection function,the detection work will consume a lot of manpower and time,and the operation efficiency is low.With the increasing demand of detection efficiency,automatic detection has become the current development trend.This thesis designs an algorithm which can detect the common defects of the contact rail surface quickly,and a neural network classifier which can realize the precise classification of the defect target.Through the computer vision technology,the maintenance and repair of the contact rail can be realized,which can greatly increase the detection efficiency of workers.In this thesis,the common defects on the surface of contact rail are taken as the research object,and the research work of defect identification and classification is carried out.The main research contents are as follows:(1)This thesis summarizes the development process of China's rail industry and the basic knowledge of contact rail,and leads to the research objectives from the domestic detection status;introduces the common methods for the rail detection,as well as the research status of defect detection algorithm based on image processing at home and abroad.(2)In order to reduce the interference caused by the camera's inherent field of view and the influence of natural factors,a series of preprocessing work is needed before the algorithm research.The method of vertical projection is used to extract the track surface area,homomorphic filtering is used to reduce noise and enhance image effect.(3)In order to improve the efficiency of rail surface defect detection,a rail surface defect detection algorithm based on frequency domain information and gradient features is proposed.Quaternion Fourier transform algorithm can calculate image saliency through image color,brightness,and motion features,analyze and obtain image spectrum and phase;calculate and build corresponding defect target saliency map in the airspace,and get defect area;and enhance the significant image effect by combining image gradient algorithm and quaternion Fourier transform,and finally complete the detection The fusion algorithm is verified by experiments.(4)Aiming at the extracted defect target,the radial basis function neural network classifier is used to classify the defect target to achieve the purpose of batch,fast and automatic defect recognition.This thesis introduces the commonly used image feature information and calculates the feature value for different types of defects.It uses Relief-F algorithm to calculate and compare the weight of the feature value,and selects the influence weight of the feature value on defect classification.The feature vector is composed of the feature values which have great difference in parameters between different defects and have certain universality in the same kind of defects.(5)The RBF neural network model is built and the corresponding parameters are set up,and the training error target is set up.Input a large number of training samples for learning training.After 177 iterations,the convergence accuracy is up to the requirement,and the final classifier model is obtained.In addition,20 test set classification tests are carried out by random sampling,and the defect classification results can be obtained directly by inputting the feature vector of the defect to be tested.The test results show that RBF neural network classifier can classify the defect types with high accuracy,and its comprehensive classification accuracy is 93%.And based on the actual detection needs,according to the research results of this thesis,a software of defect identification and classification of contact rail under Windows system is developed,which is used in the actual line detection work...