DEVELOPMENT OF AN INTEGRATED MODEL FOR BUILDING DETECTION AND ATTRIBUTE ESTIMATION USING OPEN-ACCESS SATELLITE DATA

Recent advances in deep learning enable automated building detection from remote sensing data. However, building detection typically requires costly high-resolution imagery, limiting scalability and applicability. This study proposes an integrated framework using 10 m resolution open-access Sentinel-2 data for detection and attribute estimation of buildings. Regions are classified based on building characteristics, and a multitask U-Net optimized for each region performs segmentation and building count estimation. Building objects are extracted first, followed by the estimation of structural type and construction year. Finally, earthquake-induced collapse simulations demonstrate the practical utility of the resulting data for urban planning and disaster prevention.