Abstract: To address the problem of insufficient detection accuracy of visual detection by visible light in surface defect identification of welded joints, which is mainly susceptible to factors, such as illumination variations, simple backgrounds, small defect sizes, and inconspicuous features, a weld surface defect detection method based on infrared imaging and deep learning was proposed. An infrared thermal imager was used to acquire temperature field information of joints of aluminum alloy by friction stir welding during the post-weld residual heat stage, and an infrared image defect dataset was established. Based on the YOLO model, classification and localization identification of weld surface defects were realized. To address the problem of difficulty in small-scale defect detection of infrared images, the model’s structure was optimized by introducing an attention mechanism, which suppressed global thermal diffusion information irrelevant to defects in the temperature field and enhanced defect-related local features. The results indicate that the proposed method achieves good performance in detection accuracy and stability, demonstrating its applicability for rapid weld surface defect detection.