Description of Invention
Landslides are natural disasters that pose significant risks to human life, infrastructure, and the environment. Mitigating these risks requires both proactive prevention measures and rapid response strategies. This work focuses on enhancing landslide detection using UAV imagery and deep learning algorithms. In our previous study, we used the U-Net model for landslide detection; however, its performance was limited by the small size of the dataset. To overcome these limitations, we manually re-annotated the images and expanded the dataset. Furthermore, we introduced TransUNet, a model that integrates the strengths of Transformers and U-Net architectures. Experimental results demonstrate that the combination of TransUNet and the expanded dataset significantly enhances detection performance, even in low-light scenarios or complex landslide contours. Developing more accurate landslide detection methods is anticipated to improve resident safety and facilitate faster disaster recovery.