Abstract: Objective Traditional X-ray testing methods have the problem that the evaluation results are easily affected by human factors. The evaluation personnel may experience fatigue after conducting the evaluation work for a long time,this paper aims to solve the problem that small defect targets are prone to be missed in weld images with complex backgrounds. Methods Median filter is used for image denoising. The gray level characteristic of the image changes violently at the beginning and ending positions of the weld seam is utilized, weld area extraction based on the gray change rate is carried out. The gray level of the image is enhanced, the gray level of each line is transformed, and the gray level difference between defect and background is increased by linearly stretching the gray level range. A window-cumulative S-T gray level transformation is proposed to further widen the “gray level disadvantage” of the defect, placing the transformed gray level of the defect at a low level, placing the gray level of the transformed background pixels in the gray fluctuating area at the median level. Results Combined with image de-noising, image segmentation, morphological processing and other technologies, the weak target of defects under the background of large gray fluctuations in the weld X-ray image is successfully extracted. The actual inspection image is tested using this algorithm. The comparison between the inspection results and the original image shows that the algorithm can accurately extract defects, which proves its effectiveness. Conclusion Based on the digital image processing algorithm, the proposed window-cumulative S-T transform method combines image de-noising, image segmentation, morphological processing and other technologies to successfully extract weak targets of defects in weld X-ray images under large gray fluctuations background.