A multi-scale transformer-enhanced YOLO framework for unified road damage detection and boundary-aware segmentation.

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• Road surface deterioration impacts transportation safety and infrastructure maintenance.
• Automated road damage analysis using computer vision is crucial for intelligent transportation systems.

Purpose of the Study:

• To propose a novel deep learning framework for multi-task road damage analysis.
• To integrate detection, classification, and segmentation into a unified architecture for pavement defect identification.

Main Methods:

• Developed a deep learning framework with multi-scale feature extraction and transformer-based contextual refinement.
• Incorporated ROI-based classification enhancement and a boundary-aware segmentation module.
• Utilized multi-scale alignment and contextual attention mechanisms for improved feature analysis.

Main Results:

• Achieved superior performance compared to conventional models in detection accuracy and segmentation quality.
• Demonstrated the effectiveness of contextual feature refinement and boundary-aware learning.
• Maintained practical inference speed for real-time applications.

Conclusions:

• The proposed framework offers a reliable and efficient solution for automated road condition monitoring.
• This approach supports intelligent infrastructure inspection and real-time road maintenance systems.
• Contributes to safer and more efficient transportation networks through advanced pavement defect analysis.