Detection of Weld Regions in X-ray Images of Thick Steel Pipes

Abstract

Since traditional detection algorithms of welding seam area have difficulties in accurately extracting the fuzzy and low-contrast welding areas in the X-ray images of thick steel pipes, this paper proposed a novel robust detection method of weld seam region based on the robust PCA model. The proposed technique can overcome the shortcomings of the traditional methods, and can accurately extract the weld regions. Firstly, a sequence of X-ray images were collected, and their spatial alignment and brightness normalization were carried out. Then, a series of background images were obtained, and these preprocessed images and a test X-ray image were combined to form an observation matrix. The robust PCA was then used to decompose the observation matrix into a low-rank and sparse image. As the uneven intensity and noise are greatly eliminated in the test images, the weld region of the test image is highlighted in the corresponding sparse image, and can be well segmented by a global threshold. The test results show that the uneven brightness distribution and noise from X-ray images of thick steel pipes are largely eliminated, and the weld seam edges and low contrast areas are also enhanced. Compared with the traditional welding area detection methods, the proposed algorithm can better detect the fuzzy and low-contrast welding regions with a higher detection sensitivity (0. 952) and accuracy (0. 989).

Keywords: thick steel pip, X-ray images, weld regions, edge detection, image pretreatment