Detection of possible landslides in post-event satellite images using color and texture

Abstract

We propose a method for the detection of possible landslide regions from single post-event satellite or aerial images, based on color and texture analysis. When presented as a landslide inventory map, the detected landslides will be useful for evaluating susceptibility maps accuracy, or even for use as one of the causative factors for the map itself. The proposed method demonstrated very promising performance, with 100% precision and recall in detecting potential landslides. The method also recorded 94.29% true negative rate when filtering out non-landslide regions, thus providing a strong platform for the detection of actual landslide regions after the incorporation of geometrical and landslide-specific features in the future.