To meet the need of the automation and intelligence of welding process, it’s very important to extract the edge of weld seam accurately for seam tracking. According to the characteristics of GMAW (gas metal arc welding), an image sensing system of weld pool region based on CCD (Charge-coupled Device) is established. An improved method of weld seam extraction is presented. Firstly, weld pool region localization method using saliency detection is proposed, and weld seam region is obtained from the right edge of weld pool, then Sobel transformation and computation model is used to extract the edge of weld seam. Experimental results show that our method can obtain a more accurate weld seam edge and cost less than other method.