A Novel Model for Instance Segmentation and Quantification of Bridge Surface Cracks-The YOLOv8-AFPN-MPD-IoU

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Surface cracks are critical indicators of infrastructure damage, necessitating reliable detection methods.
• Manual inspection of bridges for cracks is unreliable, time-consuming, and hazardous.
• Automated crack detection is essential for ensuring structural health and safety.

Purpose of the Study:

• To develop a novel deep learning model for precise instance segmentation and quantification of bridge surface cracks.
• To enhance feature fusion and geometric feature exploration for improved crack characterization.
• To provide an automated, accurate, and efficient alternative to manual bridge inspection.

Main Methods:

• Utilized YOLOv8s-Seg as the backbone for instance segmentation.
• Incorporated an Asymptotic Feature Pyramid Network (AFPN) for advanced feature fusion.
• Introduced Minimum Point Distance (MPD) as a loss function to better capture crack geometry.
• Employed middle aisle transformation with Euclidean distance for crack dimension calculation.

Main Results:

• The proposed YOLOv8s + AFPN + MPDIoU model achieved 90.7% precision, 70.4% recall, and 79.27% F1-score.
• Achieved mAP50 of 75.3% and mAP75 of 74.80%, outperforming contemporary models like YOLOv8 variants and Mask-RCNN.
• Demonstrated improvements in F1-score, mAP50, and mAP75 by at least 0.46%, 1.3%, and 1.4%, respectively.
• Measurement error margin was maintained at or below 5%.

Conclusions:

• The developed YOLOv8-AFPN-MPD-IoU model offers a robust solution for accurate bridge surface crack detection and quantification.
• This deep learning approach significantly enhances the efficiency and reliability of infrastructure assessment.
• The model provides a valuable tool for preserving structural integrity and ensuring public safety in bridges.