Two-Stage Defect Recognition of Small Diameter Pipe Weld Seams Based on Deep Learning

The welding quality of small pipes directly impacts the safety and stable operation of engineering projects. X-ray non-destructive testing is an important method for assessing the welding quality of seams. To improve the efficiency and accuracy of radiographic inspection, this paper proposes a two-stage intelligent detection method for X-ray defects in small pipe weld seams. In the first stage, a lightweight modified YOLO network is employed to segment the images and extract regions of interest (ROI) where defects are concentrated. The second stage involves feature analysis of the images combined with sparse dictionary learning. Finally, the sparse dictionary matrix is utilized to accurately classify and differentiate between circular defects, linear defects, and noise. Experimental results demonstrate that the proposed method effectively reduces the size of the network model, achieving a lightweight design, while the success rate for identifying circular defects is 0.974 and for linear defects is 0.967, thereby facilitating effective recognition of defects in small pipe weld images.