Drainage Pipe Defect Detection and Effect Analysis Based on Improved YOLOv8_DGW

This work redefined and categorized the drainage pipe defect classification method in China through the lens of computer vision interpretation, addressing the inefficiencies and excessive costs associated with existing methodologies. Based on this, we propose an improved YOLOv8_DGW defect detection model. In other words, three enhancements are implemented using the YOLOv8n detection model: first, the original C2f structure is replaced at the end of the backbone network with deformable convolutional DCNv2, which enhances the feature extraction capability of the model for irregular faulty targets; second, the global attention module is added to improve the model’s ability to extract picture features; and, finally, the original complete intersection over union (CIoU) loss function is replaced with the Wise-IoU (WIoU) loss function, substantially enhancing the model’s defect detection capabilities. According to the test findings, the YOLOv8_DGW model that has been improved and proposed in this work obtains a mean average precision (mAP) value of 76.8%, which is 5.1% higher than the previous model’s mAP value of 71.7%. Additionally, the model’s precision and recall have increased by 4.3% and 6.0%, respectively. The enhancements have improved the YOLOv8n’s capacity to identify drainage pipe flaws.