An Efficient and Uncertainty-Aware Decision Support System for Disaster Response Using Aerial Imagery.

Area of Science:

• Remote Sensing
• Disaster Management
• Computational Modeling

Background:

• Efficient search and rescue operations are critical during disasters.
• Remote sensing and computational models enable building damage assessment (BDA) using pre- and post-disaster aerial imagery.
• Existing BDA methods often prioritize accuracy over efficiency and uncertainty quantification.

Purpose of the Study:

• To propose an efficient and uncertain-aware decision support system (EUDSS) for building damage assessment.
• To integrate uncertainty quantification into BDA models for critical applications.
• To enhance the reliability of BDA by incorporating expert review for high-uncertainty cases.

Main Methods:

• Developed an efficient and uncertain-aware BDA model integrating Fourier attention and Monte Carlo Dropout for uncertainty quantification.
• Implemented a robust operation (RO) procedure for expert manual review when assessment uncertainty is high.
• Validated the system's effectiveness using a public dataset, assessing both quantitative and qualitative performance.

Main Results:

• The proposed EUDSS demonstrates efficient and accurate building damage assessment.
• Uncertainty quantification is effectively achieved, providing insights into model confidence.
• The RO procedure successfully addresses challenges posed by external factors like cloud clutter and poor illumination.

Conclusions:

• The EUDSS offers a significant advancement in building damage assessment for disaster response.
• Integrating efficiency, uncertainty quantification, and expert review enhances the robustness of BDA systems.
• The system's effectiveness was validated, achieving top performance in a relevant hackathon.