Abstract: Pore defects may occur in the welds of nuclear fuel elements, seriously affecting the quality and safety of the product. Therefore, a pore defect detection method was developed, which used Gaussian filtering, morphological operations, edge detection, contour fitting and other methods to extract the region of interest. Then, contrast stretching, grayscale compensation, bilateral filtering, adaptive threshold binarization and other methods were used to recognize the detection image. The experimental results showed that the proposed method achieved the recognition of porosity defects in weld seam radiographic images, improved the stability and accuracy of defect detection, and also provided an effective way to detect small-sized, nearly circular and similar targets in radiographic images in uneven light and dark environments.