Research on Defect Recognition Method of Digital Pipeline Negatives Based on Improved YOLO Algorithm

Pipe welding is widely used in industry. Defects such as cracking, porosity, slagging, imperfect shape, under-welding and cracking may negatively affect the strength and sealing performance of the welded structure, and the evaluation of weld defects is time-consuming and labor-intensive, so intelligent recognition of weld defects has become crucial. In this paper, the characteristics and morphology of various defects such as weld cracks, porosity, slag entrapment, imperfect shapes, and crazing are studied in detail, and algorithm optimization and model training are carried out for these characteristics. In addition, different types of welding defects commonly found on negatives were considered, including volumetric and planar defects, slag entrapment, and under-welding. Through an in-depth study of the imaging characteristics and basic properties of these defects, we have developed a comprehensive target detection system designed to automatically identify and categorize common weld defects on negatives, providing an efficient and accurate tool for pipeline weld quality control. Ultimately, this research aims to provide a complete solution for pipeline welding quality control by combining advanced computer vision and deep learning techniques, such as improved YOLO algorithm, to improve the efficiency of defect detection and identification, thereby providing the industry with a more reliable and efficient means of digitally evaluating pipeline negatives.