Research on Welding Defect Detection Based on Visual Technology

Welding quality is crucial to components' safety and lifespan, while welding defects are unavoidable due to process parameters, material properties, and manual operations. Visual inspection via computer vision and deep learning well compensates for traditional techniques' shortcomings. This paper reviews visual-based welding defect detection methods with emphasis on deep learning applications, analyzes core challenges like image quality, micro-defect detection, small samples, and unbalanced data, and concludes future research directions: multimodal fusion, better micro-defect detection, dataset improvement, and few-shot learning. Research shows that deep learning combined with vision technology boasts notable advantages in automation, accuracy, and efficiency, boosting detection quality and advancing welding inspection towards greater intelligence and automation.