Abstract: Penetration state is a critical indicator for evaluating welding quality, and it is closely related to molten pool morphology. To address the impact of operating condition fluctuations on welding quality, a quality monitoring system for the laser welding process based on molten pool profile point sets was developed. First, a laser welding experimental platform integrating welding processing, coaxial visual monitoring, and edge computing was established. Second, considering operational efficiency, a hybrid algorithm of maximum inter-class variance multi-threshold segmentation and edge detection based on a histogram pyramid was proposed to accurately extract the features of the high-temperature welding region and molten pool profile. Additionally, a multi-level representation mode based on polar coordinates was innovatively adopted to characterize the profile, thereby forming the molten pool profile point set. Finally, a deep convolutional network model VCAS-Net was constructed to achieve highly efficient and highly accurate mapping between the molten pool profile point set and penetration state. The results indicate that the model achieves an identification accuracy of 95.83%, and the processing time for a single image is 13.23 ms, which meets the requirements for high accuracy and real-time monitoring, and is suitable for industrial field applications