Research on enhancing road apparent crack detection based on the improved YOLOv8n model

Area of Science:

• Computer Vision
• Artificial Intelligence
• Civil Engineering

Background:

• Road surface crack detection is crucial for infrastructure maintenance.
• Existing algorithms struggle with variable crack scales and complex backgrounds.
• Inaccurate detection leads to maintenance inefficiencies.

Purpose of the Study:

• To enhance the YOLOv8n algorithm for improved road surface crack detection.
• To address limitations of insufficient accuracy and missed detections.
• To develop a more robust and efficient crack identification system.

Main Methods:

• Integration of Convolutional Block Attention Module (CBAM) with Cross-Stage Partial-feature fusion (C2f) module.
• Introduction of Spatial Pyramid Pooling Faster Cross-Stage Partial Channel (SPPFCSPC) for multi-scale feature extraction.
• Adoption of the Slim-Neck paradigm and Weighted Intersection over Union (WIOU) loss function.

Main Results:

• The enhanced YOLOv8n algorithm demonstrated significant improvements over the benchmark.
• Average Precision (mAP@50) increased by 1.8%, mAP@50-95 by 1.7%, and Recall by 1.8%.
• The refined model shows superior bounding box regression and gradient mitigation.

Conclusions:

• The improved YOLOv8n algorithm effectively enhances road surface crack detection accuracy.
• The modifications successfully handle target scale variability and background interference.
• This advanced algorithm meets the demands of modern road maintenance practices.