Railway Fastener Status Detection Under Non-Uniform Illumination Based On Image Processing

Railway fastener is the key component of railway track lines, its performance directly determines the safety of the railway. In recent years, with the high-speed passenger locomotives and heavy freight locomotives increased, it is more and more urgent for the modernization of railway line detection.As railway lines are in the natural environment. the railway image would be light inequality for the impact of natural light. It would reduce the detection accuracy of the fastener state. In view of the above problems, this paper researches on the detection algorithm of the railway fastener with the light inequality. The main work is as follows:Research the deficiency of common image enhancement algorithms using in the detection of the fastener state. Histogram equalization can improve the image contrast effectively. but the portion of the image would be excessive enhancement. Retinex can not handle the high light portion of the image, and its applicability is poor as it need adjust parameters for each image. The result of homomorphic filtering is not ideal as there much shadow region and the high light region in the railway image.For the above shortcomings, the empirical mode decomposition algorithm is introduced into the fastener image preprocessing in this paper.This algorithm is fully data driven and achieve the batch processing of the railway image easily as it never need adjust parameters for each image. At the same time, this algorithm computes the image data in the spatial domain avoided the blindness of image processing algorithm in the frequency domain which the gray value and its spatial position cannot be the corresponding. According to the advantage the empirical mode decomposition algorithm and histogram equalization, this paper presents an new image preprocessing algorithm for the light inequality railway image. The image is decomposed into illumination component and detail components by this algorithm, then enhance the detail component and adjust the illumination component to improve the image quality. Experiments show that the algorithm can reduce the light intensity of high light portion applicability and improve the accuracy of the detection of the fastener state.Under the base of above, combined with the features of the fastener profile, this paper presents an algorithm for the fasteners contour optimization based on contour tracing. The algorithm could remove invalid profile, reduce image noise and get the more effective HOG features. On this basis, the training sample library and the test sample library are established. Then use the linear SVM classifier to detect the fastener flaw. The experimental result shows that this algorithm could effectively improve the accuracy of the fastener state recognition.