Welding defects extraction method by fusing saliency information of mid-level and underlying level images

Abstract

X-ray non-destructive testing (NDT) technology is extensively utilized in the welding industry for the detection of weld defects. This paper proposes a novel defect segmentation algorithm to address the challenges of X-ray defect detection images, including low contrast, blurred edges, significant noise, and pronounced background variations. Traditional detection methods often struggle to extract low-contrast defects from weld images, so this approach integrates both underlying and mid-level image information to enhance accuracy. The process begins with a visual saliency model that generates a rough saliency map from underlying details. Next, a Pulse Coupled Neural Network (PCNN) is used to compute the saliency map at the mid-level. Finally, these two saliency maps are combined using a pixel-minimum method to produce the final image saliency map. Experimental results show that this method is highly accurate, broadly applicable, and capable of rapid defect extraction within the welding area.