Automatic Pipeline Defect Detection Based on An Improved YOLOV11-seg Instance Segmentation Network

In order to realize automatic pipeline defect detection, this paper proposes an improved real-time instance segmentation network based on YOLOv11-seg. The network introduces the BiPANet module, fuses spatial and multiscale features, enhances the recognition ability of small and medium targets, and improves recognition precision and recall while ensuring low model complexity. Experiments are conducted on a real scene dataset constructed based on the Chinese national standard CJJ181-2012 to verify the network's effectiveness. The results show that the model obtains mAP@50 of 87.2% and mAP@50:95 of 62% for 16 types of pipe defects under different model scales, which is a significant improvement over the original YOLOv11. It also maintains a high recognition speed of 3.4 ms, which meets the real-time requirement of automatic pipeline operation. Our network provides an effective solution for the detection of sewer defects.