SAR image automated detection of dune area

Detecting changes in desert areas is an important challenge for remote sensing images. This may concern the immediate security (when looking for lost people, for example), but also the measurements of environmental changes. In this paper we will develop a new method, based on a multi-user approach, for automatic classification of SAR images which can be used to detect dunes areas in the desert. This will provide a new approach for detecting changes.We will show here that only methods based on shape criteria can give good results, which justifies our choice. Our method is mainly based on multi-scale correlation between two images with two different acquisition dates, so that we can get better benefits of textures information, and minimize the effects due to the building of SAR images. We have selected an area corresponding to a part of the Tibesti (Sahara Desert), west from Faya-Larjeau, Chad. This region of study is providing various landscapes which represents a great interest for applying our algorithm: we can find there sand areas, with a diffuse backscattering, and areas with a lot of stone blocks that are generating specular backscattering. As a result, we show differences in accuracy by applying those works to a couple of Envisat images in our area of interest in Chad.