Weld defect detection is a key link to ensure the quality of welding. The problem of X-ray of weld defect detection has been widely studied with the rapid development of industry and urgent demand. However, because of imaging methods, influence of casting material and other objective factors, X-ray image noise background, low contrast, brightness uneven and weld edge blur, which make use of computer to weld defects automatic detection accuracy is not very ideal. Aiming at these problems, a new method is proposed to detect weld defects in this paper. Firstly, fast independent component analysis (ICA) is used to reconstruct the X-ray image background with defect. Then, the image is subtracted from its reconstructed image to obtain the difference image，and the method of threshold segmentation is used to extract the defects. Finally, further processing on the extracted results effectively reduces the false detection rate. Compared with other traditional detection algorithms, the proposed method is not sensitive to defect types, so it has better adaptability and versatility.