Automatic Coseismic Landslide Mapping by Integrating Classification and Change Detection Techniques

Automatic co-seismic landslide mapping plays a crucial role in rapidly assessing post-disaster impacts and supporting emergency response efforts. However, most remote sensing-based approaches depend on either classification or change detection alone, failing to fully exploit their complementary strengths and thus limiting mapping accuracy. To address this issue, this study proposes a novel co-seismic landslide mapping approach that integrates classification and change detection techniques. The proposed approach consists of two main stages: automatic landslide sample generation and classifier-based mapping, both integrating classification and change detection. Specifically, in the first stage, sample reliability is improved by combining coarse change detection using pre- and post-event images with coarse classification based on the post-event image. In the second stage, classification and change detection results are fused using a maximum class probability decision strategy to further enhance mapping accuracy. Experiments conducted on two datasets covering the Mainling area in China and the Hokkaido region in Japan demonstrate that the proposed method achieved superior performance compared with existing approaches.