The detection of weld defects based on the X-ray flaw detection is a key part of maintaining pipeline safety. The realization of high-precision and high-efficiency intelligent defect detection is an important aspect to promote the intelligence and modernization of nondestructive testing. At present, it is difficult to achieve high accuracy and efficiency with deep learning-based defect detection methods because they require a large number of labeled samples and are difficult to obtain. To address this problem, this article proposes an active small sample learning-based defect detection method for pipe welds. First, the defect detector is trained in a data-driven manner by extracting small sample features based on a lightweight neural network. Then, the inference of the unlabeled samples is used to calculate the detection and classification uncertainty, which could fully exploit the value samples. Finally, the network parameters are fine-tuned according to the high-value samples to obtain a high performance improvement with minimal cost. Experimental results show that the proposed method can improve the accuracy by about 8% with fewer samples and the guaranteed operational efficiency.