Laser welding has been widely used in aerospace automotive and other fields due to its advantages of high weld seam quality, high processing efficiency, and low heat-affected zone. During the welding process, fluctuations on the surface of the molten pool directly affect the quality of the weld formation, making it significant to investigate the flow behavior of the shallow melt. In this study, a small amount of aluminum oxide (Al₂O₃) particles was introduced as tracers during welding. By combining image processing and machine learning methods, the motion trajectories of surface oxides in the molten pool were extracted and analyzed, quantifying the changes in surface melt flow velocity at different penetration stages. By correlating features such as temperature, spatter, and molten pool spot frequency, the periodic variation patterns of vortices on both sides of the molten pool surface during complete penetration stage were revealed, and their respective correlations with spatter behavior and keyhole breakthrough phenomena were illustrated. This study provides valuable references for analyzing the surface flow of shallow molten pools and optimizing weld surface quality.