Study On Defect Recognition For Rail Surface Based On Image

In the safety and efficient development and construction of the train,the orbit is a weak link.Locomotives and trains run on the rails,while the rails bear the direct pressure of rolling stock and train wheels,and railway lines going through a variety of complex geographical environment,coupled with the heat bilges cold shrink factors,there's will be scars,cracks and ripples and other defects on the rails surface,even cracks inside the rails,etc.If the active rail surface of the running train defected but not be taken seriously,it is possible to lead to the internal damage of the rail,and even rail fracture,train derailment,finally affect the trains' operation safety.Therefore,it is significance to carry out defect detection and identification of the rail surface to maintain the railway line and ensure the safe operation of the train.At present,there are three mature testing methods: ultrasonic testing,eddy current testing and the visual method.These three methods mainly rely on manual inspection,which has a slow detection speed,low detection accuracy and security risks.The rail surface defects detection method based on image has the characteristics of high speed,high precision and high automation,with the improvement of computer technology and the rise of the machine vision technology,scientific research colleges and universities and research institutions began to apply image processing technique to detect rail surface defects.Under the modern background of railway high-speed development,the study on defect recognition for rail surface based on image has more realistic significance and broad market.The paper takes the rail as the research object,combined the knowledge of image processing with pattern recognition,then put forward a rail surface defects based on image recognition.To analyze the collected orbital images and determine whether there are defects,and identify the classification of defects.In this paper,the background and significance of the research are expounded firstly.And the importance of the identification of rail surface defect detection to train operation safety is stated.At the same time,the research status of track-based orbit detection at home and abroad is reviewed,and the related technology research is analyzed and summarized.Secondly,the collected orbit image is preprocessed.The histogram equalization is used to enhance the orbital image and increase the contrast between the track surface and the non-rail surface.The median filter,the adaptive Wiener filter and the mean filter are used to simulate the orbit de-noising.The adaptive Wiener filter is used to reduce the noise of the orbit image.The boundary of the track is determined by the characteristics of the pixel value and the non-orbital pixel value of the track surface,lay the foundation to the location and mark of rail surface defects.Then,finished to mark and extract the defects of the rail surface.Choose the watershed segmentation algorithm based on the image control mark to extract the defects of the track surface by MATLAB simulation tool.And the false edges such as burrs which are extracted after the defect segmentation are eliminated by the knowledge of morphological image processing.According to the separated target defects,the author makes the markup and extracts a single defect of the binary image and grayscale image.Finally,complete the extraction and selection of the defect features,and designed the orbital surface defect classifier by radial basis function neural network.Represented and described the defect features,and calculated the eigenvalues of the surface defects.Then after contrast to choose the rectangular feature,the density feature,the eccentricity feature,and the gray mean feature as the designed classifier input item.According to the selected defect characteristics and defect categories designed the classifier input and output.The training sample composed of the track surface defects is used to train the defect classifier based on the radial basis function neural network.After the training,the sample is used to test the classifier.According to MATLAB simulation results,the algorithm can accurately and quickly identify the kinds of rail surface defects.