Urban road surface crack detection based on U-net and ResNeXt network.

Area of Science:

• Civil Engineering
• Computer Science
• Artificial Intelligence

Background:

• Urban road surfaces are prone to cracks due to increased usage, posing traffic safety risks.
• Current manual road inspection methods are inefficient, inaccurate, and subjective.
• Automated detection systems are needed to improve road maintenance and safety.

Purpose of the Study:

• To develop an efficient and accurate automated road crack detection technology.
• To fuse U-net and ResNeXt networks for enhanced crack recognition.
• To provide a foundation for integrating crack detection into intelligent transportation systems.

Main Methods:

• Fusion of U-net and ResNeXt deep learning networks for image analysis.
• Development of an innovative detection technology for road surface cracks.
• Evaluation of detection performance on various crack types, including horizontal and vertical cracks.

Main Results:

• The proposed method demonstrated superior detection performance for horizontal and vertical road cracks.
• Significant overall classification performance was achieved, though recognition of block cracks requires further improvement.
• The method's peak video memory usage was controlled within 2.1GB, indicating practical applicability.
• Performance was notably superior compared to existing detection methods.

Conclusions:

• The developed U-net and ResNeXt fused network offers a highly effective solution for urban road crack detection.
• The technology provides accurate road surface crack information, facilitating timely remedial actions and predictive maintenance.
• Integration into intelligent transportation systems is feasible, supporting real-time monitoring and enhancing overall road infrastructure management.