THEORETICAL AND EXPERIMENTAL RESEARCH ON THE DEFECTS OF HOT ROLLED HEAVY RAIL

To realize surface defects on-line detection for hot rolled heavy rail, against the shortcoming of various traditional detection methods at present, a defects detection method based on computer vision is proposed. First, lighting and radiation characteristics of hot rolled heavy rail were analyzed, appropriate infrared (IR) light cut-off filter was chosen for collecting clear surface image. Then according to the relationship of columns of the pixel-lines, the relevancy mutual-proofread algorithm was used to locate the typical defects effectively. Finally, theoretical model of LSSVM algorithm based on principal component analysis algorithm was established, and applied to defect classification process. Experimental results indicate that relevancy mutual-proofread algorithm with good effect on defects location, PCA-LSSVM algorithm is accurate for defects classification, average accuracy rate reached to 84.7%; training time is short, with an average time of 28.3 seconds, better than the BPNN algorithm and SVM algorithm.