[S-RD24] Cutting-edge sensing technology applied to geology and resource exploration

[SRD24-01] An Unsupervised Domain Adaptation Method for Cross-Sensor and Cross-Region Landslide Detection

★Invited Papers

*Yulin Xu^1,3, Qingsong Xu², Chaojun Ouyang³, Youhei Kawamura¹ (1. Division of Sustainable Resources Engineering, Graduate School of Engineering, Hokkaido University, Sapporo, Japan, 2.Data Science in Earth Observation Technical, University of Munich, Munich, Germany, 3.State Key Laboratory of Mountain Hazards and Engineering Resilience, Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu, China)

Keywords:Landslide, Remote Sensing, Unsupervised Domain Adaptation

Co-seismic landslides pose severe risks to infrastructure and human safety, requiring timely and accurate detection. However, existing methods struggle with the challenges of cross-sensor and cross-region variations, including differences in spatial resolution, imaging angles, and terrain complexity. To address these challenges, this study explores an unsupervised domain adaptation (UDA) approach designed to improve landslide detection across heterogeneous remote sensing data sources. The method employs multi-level feature adaptation and domain alignment techniques to enhance model generalization, reducing the impact of domain shifts between different sensors and geological contexts. Experimental results on multi-sensor and multi-region landslide datasets demonstrate the effectiveness of the proposed approach, achieving notable improvements in detection performance over existing methods. This study highlights the potential of domain adaptation for advancing landslide detection, contributing to more robust disaster response and risk assessment across diverse environments.