A new technique for landslide mapping from a large-scale remote sensed image: A case study of Central Nepal

This paper presents a new technique for landslide mapping from large-scale Landsat8 images. The method introduces saliency enhancement to enhance the landslide regions, making the landslides salient objects in the image. Morphological operations are applied to the enhanced image to remove most background objects. Afterwards, digital elevation model is applied to further remove the ground objects of plain areas according to the height of landscape, since most landslides occur in mountainous areas. Final landslides are extracted by the proposal regions from selective search. The study area covers 2 degrees x2 degrees, making it more similar with practical cases, such as emergency response and landslide inventory mappings. The proposed method performs satisfactorily by detecting 99.1% of the landslides in the image, and obtains an overall accuracy of 99.8% in the landslides/background classification problem, which gets further validated in another Landsat8 image of a different site. The experiment shows that the proposed method is feasible for landslide detection from large-scale area, which may contribute to the further landslide-related research.