Abstract: To improve the quality of tube welding, a high-precision post-weld seam detection and segmentation method based on a laser vision system and a CNN was proposed for detecting and segmenting five types of small-scale welding defects with a height or width of less than 1 mm. The main body of the method consisted of a multi-task CNN, which received laser stripe point cloud information and could simultaneously detect, classify, and segment the weld seam regions. Compared with existing methods based on single-task CNNs, the proposed method could acquire more comprehensive information about welding defects. A laser stripe point cloud dataset of the weld seam region was generated by the constructed laser vision system, on which the proposed network was trained and evaluated. The results show that the method has an average accuracy of 99.58%, an average precision of 99.43%, and an average recall of 97.05% on the welding defect detection task. In addition, the method has an average IoU accuracy of 92.58% on the welding defect segmentation task. The test results on real workpieces show that the average segmentation error is less than 0.1 mm. The proposed method demonstrates high accuracy in both weld seam defect detection and segmentation tasks.