Abstract: The image processing and feature extraction technology of molten pool is an important part of intelligent welding quality monitoring for gas metal arc welding (GMAW) on ships. To address the unstable characteristics of large smoke and spatter during GMAW welding of ship hull plates, such as blurred image acquisition and difficult edge extraction, a fuzzy c-means clustering (FCM) based on mean shift (MS) optimization is proposed The image processing algorithm for In the optimization design of the welding dynamic visual sensing system, on the basis of maximizing the clarity of image information acquisition, the MS algorithm is used to obtain superpixel images to solve the sensitivity of the FCM algorithm to noise. At the same time, a weighted neighborhood window is introduced on the FCM algorithm to enhance the robustness of the MS-FCM algorithm, overcome the effects of smoke, spatter, arc light, noise, etc., and complete image segmentation and edge extraction Finally, four different image processing methods were designed for FCM, fuzzy c-means with spatial constraints (FCM-S), enhanced fuzzy c-means (ENFCM), and fuzzy local information c-means clustering (FLICM) algorithms. The edge segmentation effects were compared with the MS-FCM optimization model to obtain the extracted fusion widths from these methods, Verify the accuracy of extracting geometric features of the molten pool The results show that the MS-FCM algorithm can effectively suppress noise interference, smooth information, and achieve high extraction accuracy in ship welding pool image processing.