Automated crack localization for road safety using contextual u-net with spatial-channel feature integration.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Road Engineering

Background:

• Traditional crack detection methods struggle with varying conditions.
• Accurate road crack localization is vital for safety and maintenance.

Purpose of the Study:

• To develop an automated road crack localization framework.
• To improve crack detection accuracy using deep learning.

Main Methods:

• Utilized a contextual U-Net deep learning model for pixel-level segmentation.
• Employed an EfficientNet encoder for feature extraction.
• Integrated a hierarchical attention mechanism for adaptive feature focus.

Main Results:

• The framework accurately localizes cracks by analyzing spatial and channel-wise features.
• The attention mechanism enhances focus on relevant crack features at multiple scales.
• Demonstrated effectiveness on benchmarked and custom datasets.

Conclusions:

• The proposed deep learning framework offers a robust solution for automated road crack detection.
• Contextual U-Net with attention improves localization accuracy under diverse conditions.
• This approach enhances road safety and maintenance efficiency.