Research on Automated Post-Earthquake Building Damage Assessment Method

Abstract. This study addresses the challenges of low efficiency and limited scalability in traditional post-earthquake building damage assessment methods by proposing an automated assessment framework. The approach combines an improved Segformer model, a grid-based quantitative statistical method, and LLM-driven report generation for scalable and accurate structural damage assessment from high-resolution satellite imagery. First, an improved Segformer model is developed to extract and compare pre- and post-earthquake building footprints, with optimized feature fusion and training strategies tailored for post-disaster scenarios. The model effectively detects changes in building footprints under complex conditions. Second, the study introduces a grid-based quantitative statistical method that divides the affected area into uniform cells, within which damage is classified into four severity categories.To further streamline the process, the workflow is integrated into Dify, allowing for automated processing, interpretation, and report generation via LLMs. This integration enables quick and consistent delivery of actionable insights to decision-makers, reducing the need for human intervention. The method was validated using data from the 2025 Shigatse earthquake, where the model achieved a MIOU over 86% for building footprint extraction, and the damage classification showed strong alignment with ground-truth data.This study provides an efficient and scalable solution for post-earthquake building damage assessment, significantly enhancing disaster response capabilities and urban resilience planning.