Abstract:
Coseismic landslide mapping plays a crucial role in emergency response and disaster assessment. To improve landslide identification, this paper proposes a novel and enhanced model, MultiU-EGANet. The model is built upon the U-Net architecture as the baseline, with the introduction of the MultiRes module to extract feature information across multiple scales. Additionally, the Edge-Guided Attention (EGA) module is incorporated to enhance the delineation of landslide boundaries using the Laplace operator, thereby improving the segmentation accuracy at the boundaries. A composite loss function, combining Dice loss and Focal loss, is designed to further enhance the model's robustness. Using landslide data from the Jiuzhaigou area, experimental results demonstrate that the proposed model significantly improves landslide identification accuracy compared to the baseline model. Furthermore, comparative experiments conducted with landslide data from Hokkaido show that the proposed method outperforms existing models in landslide identification tasks, with F1 scores increasing by 33.31%, 5.45%, 2.31%, and 2.18%, respectively. These results validate the effectiveness of the proposed method for coseismic landslide identification.
点击查看大图
下载:
