Post Disaster Damage Assessment Using Ultra-High-Resolution Aerial Imagery with Semi-Supervised Transformers

Area of Science:

• Disaster Management
• Artificial Intelligence
• Remote Sensing

Background:

• Preliminary damage assessments (PDA) are crucial for disaster recovery.
• Traditional door-to-door inspections are slow and inefficient.
• Existing automated PDA methods using satellite imagery and CNNs lack sufficient accuracy.

Purpose of the Study:

• To develop a more accurate and efficient automated PDA framework.
• To improve damage level predictions for entire buildings using novel data and models.

Main Methods:

• Utilized ultra-high-resolution aerial (UHRA) images.
• Employed state-of-the-art transformer models.
• Implemented semi-supervised learning with large unlabeled datasets.

Main Results:

• Semi-supervised transformer models achieved superior accuracy and generalization.
• The proposed framework outperformed existing PDA methods.
• UHRA images combined with transformers overcame limitations of satellite imagery and CNNs.

Conclusions:

• The novel PDA framework offers more accurate and efficient building damage assessments.
• Semi-supervised learning with UHRA images is key to improving disaster recovery.
• This approach enhances governmental resource allocation post-disaster.