Aiming at the current situation of manual visual inspection for X-ray weld defect detection and rating, an image preprocessing-based weld defect detection and automatic rating system is designed. The preprocessing stage introduces a bilateral filtering algorithm for the original image noise problem to reduce the impact of noise; the improved CLAHE algorithm is proposed for the phenomenon of low contrast and low gray value, which effectively improves the image gray distribution, enhances the contrast between the weld region and the background region, and highlights the key feature information; the preprocessed image is trained for deep learning, and the GUI interface calls the training results to display the defect type The GUI interface calls the training results to display the defect types and automatically rates the defects according to the rating criteria. After the experiment, the interface can quickly recognize the defect types and display the rating results after preprocessing.