Enhancing bridge damage detection with Mamba-Enhanced HRNet for semantic segmentation.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Urbanization increases the importance of bridge structural health for public safety.
• Existing methods for bridge damage detection require enhancement for accuracy and efficiency.

Purpose of the Study:

• To propose Mamba-Enhanced HRNet, a novel deep learning model for effective bridge damage detection.
• To improve the semantic segmentation of bridge damages by integrating multi-resolution analysis and visual state space models.

Main Methods:

• Developed Mamba-Enhanced HRNet by integrating HRNet's parallel design with VMamba's visual state space model.
• Replaced residual convolutional blocks with VSS blocks and convolution for enhanced global context capture.
• Created an extensive dataset featuring diverse bridge damage types for model training and evaluation.

Main Results:

• Mamba-Enhanced HRNet achieved a Mean Intersection over Union (Mean IoU) score of 0.963.
• The model demonstrated superior performance in bridge damage semantic segmentation compared to other state-of-the-art models.
• The integration of VSS blocks improved the model's ability to capture global contextual information efficiently.

Conclusions:

• Mamba-Enhanced HRNet offers a significant advancement in automated bridge damage detection.
• The model's enhanced capability in semantic segmentation contributes to improved structural health monitoring and public safety.
• This approach provides a computationally efficient and highly accurate solution for identifying various bridge damages.