Multi-Modal Attention for Automated Disaster Damage Assessment Using Remote Sensing Imagery and Deep Learning

The paper presents a novel framework for automated disaster damage assessment using remote sensing imagery and deep learning, specifically employing a lightweight ConvNeXT-Tiny backbone. It introduces a multi-modal attention mechanism that fuses bi-temporal features to classify building damage into four categories with a reported classification accuracy of 94.90% on a large-scale dataset. This approach enhances the speed and accuracy of damage assessments, providing a scalable solution for emergency responders to prioritize interventions effectively.