ViGeo-Fuse: A multimodal framework integrating computer vision and geospatial ML for post-earthquake building damage classification

Rapid and reliable assessment of post-earthquake building damage is critical for emergency response, yet existing approaches often rely solely on visual information or are limited by delayed remote sensing data. This study presents ViGeo-Fuse, an application-oriented multimodal framework integrating vision-based deep learning with hazard-informed geospatial prior knowledge for damage classification under real-world constraints. The main contributions include: (1) the construction of a georeferenced dataset comprising over 10,000 ground-level images from post-earthquake field surveys of the 2023 Jishishan earthquake; (2) the development of a confidence-weighted decision-level fusion strategy integrating a vision-based model (VGG16) and a geospatial machine learning model (Random Forest); and (3) the validation of the proposed framework, achieving an overall accuracy of 77.92% and demonstrating competitive performance, improved robustness over single-modality approaches. The results demonstrate the robustness of the framework under data-constrained conditions, providing a practical basis for post-earthquake building damage assessment and supporting emergency response.