Regression Analysis and Recognization of Weld Images Based on Texture Features

The deep research is done in this paper for the images with weld-seam through the reflect arc light from reflector beamed to the surface of well-meant. Firstly, according to the method of texture, the coordinates of the weld center in the small area are determined. Secondly, the types of the weld in the images are determined by using the rate of change of 1^st derivative changes. Thirdly, recognition methods of weld-shapes of the combined-types such as straight line, arc, ploy line of the weld from CCD are put forward by using the regression analysis. Finally through our experiments, the weld can be recognized quickly by regression equation and the time to obtain trajectories has been lessened, which provided basic guarantees for real-time tracking of the whole system.