Research on Improved YOLOv5 Pipeline Defect Detection Algorithm

Abstract

The main problem addressed in this research is the detection of three types of defects in underground drainage pipelines: gravel intrusion, obstacles, and foreign objects. To tackle this, improvements have been made by incorporating the YOLOv5 algorithm with the attention mechanisms known as the enhanced convolutional block attention module (ECBAM) and switchable atrous convolution (SAC) module. By introducing the redesigned CBAM mechanism, both channel and spatial attention can take the original image as input, enhancing the model’s focus on important features while suppressing irrelevant ones. Additionally, integrating the dilated convolution module into the original 3-layer convolution (C3) module expands the model’s receptive field and improves its perception capabilities. Finally, the smoothed intersection over union (SIOU) metric enables a more comprehensive evaluation of the matching between predicted and ground truth bounding boxes, providing more accurate guidance for model optimization. The improved algorithm achieved an mean average precision (mAP) of 64.49% in identifying three types of defects in underground drainage pipes, representing a 5.27% increase compared to the original algorithm. This indicates that the improved algorithm showed a certain enhancement in the accuracy of identifying defects in underground drainage pipelines, and it has been applied in real-world conditions for detecting underground drainage pipeline defects.