Crack Detection and Comparison Study Based on Faster R-CNN and Mask R-CNN.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Road Infrastructure Maintenance

Background:

• Intelligent operation and maintenance of infrastructure rely on accurate crack detection.
• Computer vision and deep learning have advanced road pavement crack recognition.
• Convolutional neural networks show superior performance in crack identification.

Purpose of the Study:

• To investigate deep learning for intelligent road crack detection.
• To compare and analyze Faster R-CNN and Mask R-CNN for this task.
• To evaluate the effectiveness of a joint training strategy.

Main Methods:

• Deep learning algorithms, specifically Faster R-CNN and Mask R-CNN, were employed.
• A comparative analysis of these models was conducted.
• A joint training strategy was implemented and assessed.

Main Results:

• The joint training strategy proved effective for both Faster R-CNN and Mask R-CNN.
• Successful crack detection was achieved with over 130 images.
• The proposed method outperformed YOLOv3 in crack detection.
• Mask R-CNN's bounding box accuracy degraded with joint training.

Conclusions:

• Deep learning, particularly Faster R-CNN and Mask R-CNN with joint training, offers an effective solution for intelligent road crack detection.
• This approach is viable even with limited datasets.
• Further optimization is needed to address Mask R-CNN's bounding box performance limitations.