Underwater welding seam image is subjected to effect by noise,such as underwater environment,arc,spatter,sediment and so on,making the image poor quality.If classics edge detection algorithm is used to extract underwater welding seam edge,the result isn’〖KG-*2〗t satisfactory.Canny algorithm was used as the foundation,a auto-adaptive double-threshold edge detection algorithm was proposed based on multiscale wavelet transform.In the algorithm,multiscale wavelet transform was used,cubic B-spline was used as wavelet function,double linear interpolation was used to suppress non-maximum,auto-adaptive double-threshold generated by using OTSU method was used to restrain noise and remove pseudo boundary.The experimental results show that this method is better than the tradition Canny algorithm,the edge detection is more accurate and complete,moreover the noise is restrained effectively.The validity of the algorithm is verified.