Image Recognition Based On Object Contour Feature And Its Application In Train Bogie Fault Diagnosis

The train, including fast train and CRH train, is one of main transports of passengertraffic and freight transport. With the development of modern industry, science andtechnology, the size and complexity is increasing day by day. Thus, if the systems areinterrupted by faults, heave losses in manpower and material resources will be suffered. Solarge numbers of manpower, material resources and financial is devoted to the study of thetrain bogie’s fault diagnosis and condition monitoring in many countries. The faultrecognition based on optical image has become one of the most important researchdomains. Object recognition is always one of the hot topics and difficulties in the field ofimage processing, pattern recognition and computer vision, and it is widely used in the areaof the daily lives, industrial applications and military activities. In this dissertation, thestudy of train bogie’s fault recognition, based on the shape representation and matching ofobject contour, is put forward. The main contents are listed as follows.Firstly, the image pre-processing for optical image of train bogie is studies. In thispaper, we systematically summarize the research achievements and the current situation ofresearch related to image edge detection technology at home and abroad. The edgedetection algorithms with differential operators of one order and second order areintroduced includes Roberts, Sobel, Prewitt and Canny operator. Then the edge detectionalgorithm of multi-scale wavelet transform is presented, and we introduce the basicprinciple of wavelet transform and the procedure of wavelet edge detection algorithm.Secondly, the object recognition algorithm based on PCA-SC global feature is studied.In this section, we propose a new descriptor, called PCA-SC descriptor, which appliesPrincipal Components Analysis (PCA) algorithm to reduce the dimensions of featurematrix formed by Shape Contexts. In the proposed PCA-SC algorithm, we build acovariance matrix, and reduce its dimensions according to the size of eigen value. PCA-SCdescriptor can not only remove noise interference and improve the recognition accuracy, but also enhance the matching efficiency. The experimental results of MNIST databaseindicate that the PCA-SC descriptor outperforms previous SC algorithm. Furthermore, theanti-noise performance becomes stronger.Thirdly, the object recognition algorithm based on hierarchical representation andflexible matching is studied. Efficient object contour segment and shape matching methodare the critical problem to describe the local features of objects and similarityrepresentation. In this section, a contour description algorithm called hierarchicaldescription, based on cognitive psychology, is put forward to solve the improper contoursegments obtained by the existing recognition methods. Meanwhile, a flexible matchingmethod, based on the contour segment’s geometric feature of length, camber andbend-ability, is proposed. First, the whole contour is divided into several contour segmentsby the corners. Then the valuation scale is put forward to combine these contour segmentsinto several contour feature segments. Finally, the similarity of different contour featuresegments is computed in flexible matching mode to get the best recognized result. Theexperimental results of MPEG-7and Kimia database indicate that this algorithm has greatadvantage over recently published algorithms, especially for the objects with partialocclusion.Finally, the proposed method is used in the application of detection and recognition offault train bogie and validating the feasibility of the method. The main objective of thisrecognition system is to recognize the important and fault parts based on the movingbogie’s optical image with standard image. The experimental results indicate that theperformance of the presented method is correct and efficient, and the ratio of common faultrecognition is high.