Shuffle Attention-Based Pavement-Sealed Crack Distress Detection.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Pavement performance stability is crucial for infrastructure longevity.
• Accurate detection of pavement-sealed cracks is essential for timely maintenance.
• Existing methods may lack the precision needed for complex crack patterns.

Purpose of the Study:

• To introduce a novel shuffle attention-based approach for pavement-sealed crack detection.
• To improve the accuracy and efficiency of identifying pavement cracks.
• To ensure the long-term stability and performance of road infrastructure.

Main Methods:

• Integration of a shuffle attention module within feature extraction and detection networks.
• Utilizing a two-way detection head with multi-scale feature fusion.
• Employing a Wise Intersection over Union loss function for dynamic gradient adjustment.

Main Results:

• Achieved a mean Average Precision (mAP@0.5) of 98.02%.
• Recorded a Recall of 0.9768 and an F1-score of 0.9680.
• Demonstrated superior performance compared to existing one-stage state-of-the-art methods.

Conclusions:

• The proposed shuffle attention-based method significantly enhances pavement-sealed crack detection accuracy.
• The Wise Intersection over Union loss function improves bounding box fitting and coverage.
• This approach offers a promising solution for maintaining pavement integrity and performance.