[A-CG41] Satellite Earth Environment Observation

[ACG41-P01] SAR-Optical Data Fusion for Post-earthquake Building Damage Assessment

Shao-Ming Lu¹, *Szu-Yun Lin¹ (1.National Taiwan University)

Keywords:Disaster Management, Building Damage Assessment, Deep Learning, Data Fusion, Satellite Imagery, Synthetic-Aperture Radar

As global climate change intensifies, the frequency and severity of natural disasters continue to rise, making timely assessments and responses critical challenges. Traditionally, post-disaster building damage assessment has been time-consuming, labor-intensive, and fraught with considerable risks. In recent years, rapid advances in computer vision have enabled deep learning techniques to be widely adopted across various fields, significantly reducing the time required for damage interpretation and markedly improving accuracy. Satellite imagery, with its extensive coverage and timely availability, provides an initial assessment method during the early stages of a disaster. Although many studies have applied deep learning to post-disaster assessment classification tasks, the data diversity bottleneck limits the effectiveness of training with single-modality inputs. Moreover, adverse weather conditions or cloud cover may lead to data deficiencies.
To enhance a model’s ability to learn from multiple data sources, this study applies deep learning to both optical and SAR imagery for damage assessment of individual buildings. By employing a multimodal fusion strategy, we integrate spatially complementary information from these two data types, thereby mitigating issues arising from differences in illumination, weather conditions, and material properties. Furthermore, the 2021 Haiti earthquake and the 2023 Turkey earthquake are used as case studies, wherein post-disaster imagery is preprocessed and analyzed using the proposed multimodal model. This approach not only improves the practical effectiveness of building damage assessments but also enhances overall efficiency.