Silicon steel sheets are widely used in the production of motor stator cores due to their good magnetic properties. The molten pool is an important physical phenomenon in the welding process, and its dynamic behavior affects the welding quality. This paper studies the welding molten pool, and studies the influence of welding parameters on the molten pool and the temperature field changes of the molten pool through a combination of visual monitoring of the welding process and numerical simulation. At the same time, the U-Net network model has been improved, which improves network performance while achieving network lightweighting, and can accurately segment the melt pool and extract features. The average intersection ratio and average pixel accuracy reached 95.91% and 98.59% respectively, both of which were better than the original model. Numerical simulation results show that the molten pool area and welding depth are positively correlated with the molten pool temperature, and monitoring the molten pool area can better reflect the changes in welding depth. This facilitates online monitoring of the welding process and prediction of welding depth.