Image-To-Insight: A novel workflow for converting post-disaster imagery of historic masonry structures into actionable data

Post-disaster damage assessment of historic masonry buildings presents unique challenges due to the complexity and variability of damage types, as well as the need for rapid and detailed evaluations based on reconnaissance data to guide recovery efforts. This paper introduces a novel methodology that integrates automated 2D image segmentation with 3D photogrammetric point clouds to detect, quantify, and visualize multi-type and multi-scale damage in historic masonry buildings. Using a Mask R-CNN-based segmentation model trained on a dataset of 6,000 images collected after the 2020 Beirut explosion, the methodology achieves accurate damage detection with an overall model test accuracy of 88%. The proposed 2D-to-3D segmentation process transfers image-based segmentation masks onto 3D point clouds, automating the spatially precise classification of 7 different damage types. The segmented 3D models are further processed to extract structured data, including dimensions of individual damage instances, enabling the generation of damage inventories critical for restoration planning. Future applications of this methodology could extend to other historic and modern structures by incorporating additional datasets through transfer learning. By integrating advanced computer vision with photogrammetric modeling, this research contributes a transformative tool for post-disaster resilience planning, bridging the gap between manual assessments and automated workflows for recovery recommendations.