Superpixel-based automatic image recognition for landslide deformation areas

Obtaining continuous landslide deformation information is important for analyses of landslide processes. This paper proposes a landslide deformation area image recognition method. Using this method, the landslide area Can be automatically identified, and continuous landslide deformation data can be obtained. This novel method is implemented using Python and the OpenCV open source libraries, and it is validated using video data of rainfall-induced landslide experiments. The results indicate that the image recognition method can identify the landslide features in video monitoring images with high accuracy. An analysis of the error in the recognition results shows that the environmental changes in light intensity and specular reflections are the main causes of the recognition errors. The new image recognition method is straightforward and can be applied to bare slopes, especially man-made slopes.