Abstract: This paper focused on the small sample problem in weld defect detection, covering multiple types of data such as X-ray inspection images, visual inspection images, and ultrasonic inspection signals and images. From the two dimensions of dataset construction and algorithm optimization, the research progress on the small sample problem in weld defect detection was systematically reviewed. At the dataset level, techniques including geometric transformation augmentation, generative data augmentation, and image quality enhancement were summarized, which alleviated the small sample constraints through sample expansion and quality optimization. At the algorithm level, the applications of transfer learning and meta-learning were mainly analyzed. Transfer learning adapts pretrained parameters from large-scale general datasets to the detection task, while meta-learning strengthens the rapid adaptation ability of the model to small samples by enabling the model to “learn how to learn”. The application effects and limitations of various methods were compared, pointing out that data augmentation techniques have difficulty in enriching the semantic diversity of defects, and the performance of transfer learning is limited in scenarios with poor data quality. Meanwhile, the excellent application effects shown by meta-learning technology in the field of defect detection were summarized. This technology is currently less studied and applied in the field of weld defect detection, but it has great development potential. This paper provides a reference for subsequent research and engineering applications of the small sample problem in weld defect detection.