Laser welding is divided into two categories: laser heat conduction welding and laser deep penetration welding according to the power density. It has different weld seam characteristics. Due to the different welding positions and styles of the products, it is impossible to use traditional optical detection to distinguish the annular distribution of the weld seam, the appearance of the finished product such as no welding, etc., so the production line is finally inspected manually, which often causes the welding quality to be uneven.By using the Classification and Segmentation technology of Solomon SolVision AI image platform, AI models can be trained with weld feature images to identify welding power and no welding defects, and the number and distribution of fish scale patterns of weld seams can be accurately detected through deep learning.