Abstract: To enhance welding quality and production efficiency, this paper proposes an automated welding defect detection method based on machine vision. This method realizes intelligent detection and quantitative evaluation of welding defects by constructing a system architecture that includes core modules such as image acquisition, preprocessing, and defect recognition. Firstly, in the image processing stage, median filtering and Gaussian filtering algorithms are comprehensively applied to effectively suppress noise, and through steps such as grayscale conversion, binarization, and morphological processing, the distinguishability of weld seam and defect features is significantly enhanced. Secondly, by combining edge detection technology with defect area quantitative analysis, common defect types in weld seams such as pores, cracks, and incomplete fusion are accurately identified, providing reliable data support for welding quality control. Finally, experimental verification is conducted, and the experimental results show that the detection accuracy of this method for typical welding defects such as cracks, pores, and incomplete fusion reaches a high level, with the average error between the measurement results and the manual measurement values controlled within 0.1 cm, fully meeting the actual application requirements of industrial production.