A robust self-supervised approach for fine-grained crack detection in concrete structures.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Concrete structures are prone to deterioration from fine-grained cracks, compromising structural integrity.
• Existing computer vision methods, including deep convolutional neural networks and transformers, struggle with precise localization of these fine cracks.
• Automated crack detection is crucial for timely maintenance and extending the lifespan of concrete infrastructure.

Purpose of the Study:

• To develop an advanced computer vision approach for accurate detection and segmentation of fine-grained cracks in concrete structures.
• To improve upon existing methods that face challenges in localizing subtle crack features.
• To enhance the efficiency and accuracy of automated structural health monitoring systems.

Main Methods:

• A novel self-supervised 'you only look once' (SS-YOLO) approach based on the YOLOv8 model.
• Integration of Convolutional Block Attention (CBAM) and Gaussian Adaptive Weight Distribution Multi-Head Self-Attention (GAWD-MHSA) modules for enhanced feature identification.
• Application of curriculum learning-based self-supervised pseudo-labeling (CL-SSPL) to improve performance on limited datasets.

Main Results:

• Achieved a mean average precision (mAP) of over 90.01% and an F1 score of 87%.
• Demonstrated significant improvements of at least 2.62% in mAP and 4.40% in F1 score compared to existing methods across three datasets.
• Exhibited a reduction in inference time by 2 ms per image, indicating increased efficiency.

Conclusions:

• The proposed SS-YOLO method effectively addresses the challenges of fine-grained crack detection and segmentation in concrete structures.
• The integration of attention mechanisms and pseudo-labeling significantly boosts detection accuracy and robustness.
• This approach offers a promising solution for automated structural health monitoring, enhancing the reliability and longevity of concrete infrastructure.