Abstract: 　　
　　Aiming at the problems of strong subjectivity, low efficiency, and poor data traceability in the traditional manual film evaluation method for defects in welded joints of pressure equipment, an intelligent detection and rating method integrating deep learning target detection technology and the NB/T47013 standard is proposed. A welding defect dataset is constructed and the YOLOv8s model (a lightweight version of the YOLO series target detection algorithms, adopting the C2f module and SPPF spatial pyramid pooling structure, which has both advantages of detection accuracy and inference speed) is trained. The sliding window inference and coordinate mapping strategy are used to realize full-region defect recognition and localization of high-resolution films. Combined with the DPI (dots per inch, a core parameter for measuring image digitization accuracy, used to establish the conversion relationship between pixel size and physical size) pixel-to-physical size conversion and the rating rule engine adaptive to working condition parameters, the accurate calculation of defect parameters and standardized quality rating are completed, and a structured detection report is finally generated. Experimental results show that the method can effectively identify various defect types such as porosity, cracks, and incomplete penetration. The consistency between the rating results and the standard reaches 95%, and the detection efficiency is 5 times higher than that of manual film evaluation. It provides an efficient, objective, and traceable quality assurance scheme for the safe operation of pressure equipment.