Based on machine vision, the image processing and feature extraction of laser welding pool were studied in this paper. The results show that the monitoring results of optical fiber laser deep penetration welding pool are analyzed, including penetration status, weld width deviation tracking monitoring and line monitoring. Under the condition of welding speed and different welding power, the welding monitoring system can realize the real-time monitoring of optical fiber laser deep penetration welding pool, and its accuracy can meet the requirements of laser welding. In the process of deep penetration welding, the change of welding speed has a significant effect on the welding process. The penetration phenomenon can change the heat dissipation condition and cause the difference of the geometrical characteristic quantity of the molten pool. Determine the coefficient of change of weld width in advance, and judge whether there is penetration in the welding process. Track and monitor the distribution range of sub-pixel corner points and the trend of weld pool characteristics, and effectively judge the welding through the phenomenon of weld deviation.