Internal defects are the main factor which affect the quality of welded structures. In order to analyze the internal defects of the weld, image data with high sensitivity and sharp edges is needed to acquire. Detection method of noised color image edge based on morphology is proposed according to the characteristics of the welded defects and specialty of sound image. This paper achieves the effective segmentation of the defect information through filter process of original image, distinguishes the characteristics of defects by mathematical statistical method based on characteristics of defects. Finally, experiment proves that this method have a relatively good result for correct identification of the defects.