Abstract: During the development of welding from automation to intelligence, it requires real-time acquisition of clear images of the welding process for monitoring the weld pool morphology, tracking the weld seam, and optimizing the welding parameters and trajectory. However, due to the interference of strong arc light, it is difficult for conventional vision sensors with a dynamic range of less than 100 dB to clearly capture key welding information such as the weld pool, tungsten pole and weld seam. Therefore, a wide-band pass optical filter and a neutral density filter were added on the basis of the vision sensor with 86 dB dynamic range, and a double Gaussian filtering algorithm was combined to capture the welding process image.According to the distribution of the arc light spectrum, a wide-band optical band-pass filter in 350-670 nm was selected,which not only filtered out the strong arc light radiation in the infrared band, but also overcomed the disadvantage of single-color image by using the narrow-band pass optical filter. The results show that when the double Gaussian filtering algorithm is used to enhance the contrast between the weld pool and the weld seam region, the details of the weld seam and weld pool edge of the processed welding image are more visible, and the information in the image is clearer.