Abstract: Automatic detection and recognition of weld defects is one of the hot spots in nondestructive testing. In this paper, firstly, a smooth template was constructed, the original image was filtered by median filter, and then the ideal weld image was constructed. Secondly, the image subtraction operation was performed, and the current position was marked as a suspicious defect when the gray connectivity exceeded a given threshold value of T, then automatic detection of weld defects was realized. Four suspicious defects were obtained after automatic detection, the characteristic parameters of all suspicious defects were calculated, and all suspicious defects were determined as gas pores after qualitative analysis. Finally, the defect list was obtained, and the correspondence between the defect list and the two value image of gas pores was established.