Design And Research Of Recognition System On Railway Surface Defects

With the development of the railway, defect inspection of rails becomes more and more important to ensure the safety. In order to overcome the disadvantages of manual inspections and to adapt the trend of automatic inspections, this paper advances a new inspection method of rail surface defects based upon image processing, pattern recognition and machine vision. The advantage of it is not only decreasing the effect of environmental factor and economizing on manpower, but also advancing the detection rate of speed and precision.It has high potential applicationsThe paper is composed of a hardware part and a software part.The first part mainly designs the hardware frame of the system using for reference universal the model of machine vision system. It aims at the system precision's demand and chooses the right components and computers each component' parameters. It also analyses the influence of various speed to thepicture quality, and then designs the project-changing dynamically the linerate of the camera to solve the problem so that images have the same resolutions intwo dimensions. Then it has carried on the experiment to verify the possibility of this project. And it acquires a great lot of images by using the hardware platform.The second part is mainly about image processing and pattern recognition. According to the surface defect image's traits and using digital image processing it abstracts the defects from the rail images, describes the feature of them, and classifies them. The key steps include pretreatment, rail location determination, edge detection, edge link, defect orientation, feature extraction and pattern classification. Under-mentioned arithmetic has a great influence on inspecting and analyzing defects exactly.1. In order to avoid the disturbance of the macadam and others near by the rail, rail region is oriented by using vertical edge detection. The operation can intensify rail's borderline and cut it down by projection.2. The edge of the defect is abnormity. In order to abstract the defect withhigh precision, improved Sobel method and canny method are applied to carry on edge detection. The results are effective and can preserve edge information better.3. Improved edge growing arithmetic is advanced to link discontinuous edge, which increases connectedness.4. It describes the features after orientation to defects, and chooses aspect ratio and compactness as the input of Learning Vector Quantization Neural Network, and then computers the parameters of classify system. Flaws and scars are classified exactly.At last the article provides the results of image processing and pattern recognition in allusion to different types of defects, which prove that the system could inspect rail surface defects exactly and have stated applicability.