Abstract: To address the current situation of low accuracy rate of small-diameter tube welding image defect detection, by combining image feature analysis and sparse dictionary learning, a small-diameter tube welding defect detection algorithm based on image segmentation is proposed. Firstly, using two-step image segmentation way acquires the region of interest which is in small-diameter tube welding image. Secondly, the suspected defect region is obtained by extracting welding defect. Finally, we propose a mathematical model of the dictionary matrix of small-diameter tube welding defects with the objective of minimizing correlations between different types of atoms and solve it by using K-SVD algorithm. After that, the dictionary matrix is used to classify circular defects, strip defects and noise. To improve the real-time performance of the system, we use parallel programming to accelerate the image segmentation algorithm. The results show that the recognition rate of the proposed method is 0.974 for circular defects and 0.967 for strip defects, and the recognition speed is fast, which enables the effective recognition of defects in small-diameter tube welding image.