Related Experiment Video
Updated: Jul 5, 2025
11:34
Subsurface Defect Localization by Structured Heating Using Laser Projected Photothermal Thermography
Published on: May 15, 2017
11.1K
Research on surface defect detection algorithm of pipeline weld based on YOLOv7
1School of Material Science and Engineering, Xi'an Shiyou University, No. 18, East Section of Electronic Second Road, Xi'an, 710065, Shaanxi, China. xqxu@xsyu.edu.cn.
Scientific Reports
|January 22, 2024
Summary
This study introduces an improved YOLOv7 model for pipeline weld defect detection, significantly reducing missed detections and improving accuracy. The enhanced model achieves a 78.6% mAP, outperforming previous methods.
Area of Science:
- Mechanical Engineering
- Computer Vision
- Materials Science
Background:
- Traditional weld surface defect detection methods suffer from low accuracy and high leakage rates.
- Existing deep learning models struggle with feature extraction and interference in weld defect images.
Purpose of the Study:
- To enhance the accuracy and reduce the leakage rate of pipeline weld surface defect detection.
- To improve the feature extraction capabilities and target representation in weld defect identification.
Main Methods:
- An improved YOLOv7 model incorporating a Le-HorBlock module for second-order spatial interaction.
- Integration of Coordinate Attention (CoordAtt) block to enhance feature representation and suppress interference.
- Replacement of CIoU loss with SIoU loss for optimized convergence and reduced model freedom.
- Utilized a new large-scale dataset of 2000 pipeline weld defect images.
Main Results:
- The improved YOLOv7 model demonstrated a significant reduction in the missed detection rate compared to the original network.
- Achieved a mean Average Precision (mAP@80.5) of 78.6%, a 15.9% improvement over the original YOLOv7 model.
- The enhanced model outperformed both the original YOLOv7 and other classical target detection networks.
Conclusions:
- The proposed improved YOLOv7 model effectively enhances pipeline weld surface defect detection accuracy.
- The integration of Le-HorBlock, CoordAtt, and SIoU loss contributes to superior feature extraction and detection performance.
- This advanced model offers a more robust solution for identifying weld defects in industrial applications.
