Abstract: To address the limitations of existing weld defect detection methods, this study integrated magnetic particle testing principles with image processing technologies, focusing on the acquisition and processing of magnetic particle indications as well as defect extraction and recognition. A machine vision-based autonomous identification system for weld magnetic particle testing defects was proposed and experimentally validated. The results demonstrated that grayscale conversion and enhancement processing significantly improved crack clarity in images while reducing interference factors. Median filtering outperformed mean and Gaussian filtering in noise reduction, exhibiting stronger contrast. Images processed using the maximum entropy method facilitated feature extraction. The system innovatively combined the Sobel algorithm and Canny operator, integrating crack and porosity identification results into a designed GUI human-machine interaction system, achieving fully autonomous defect recognition in images. This approach effectively combined magnetic particle inspection with machine vision technology, establishing a solid foundation for intelligent welding inspection.