Diversity of settlement categories in very high resolution SAR images

In this paper we propose to identify the number of urban categories that can be retrieved from very high resolution TerraSAR-X data. For this task a semi-automated procedure was built in order to search in large Earth Observation database for similar sub-images (i.e. patches) and group them in the same semantic category. The dataset consists of 39 scenes over the world: 18 scenes in Asia, 15 scenes in Europe, 5 scenes in North and South America, and 1 scene in Africa. These scenes are grouped in three different types of collections (a total of 16 collections are generated) and the semantic categories in each collection are retrieved. A total of 320 categories are identified and those that contain man-made structures in the urban area are kept and the rest of categories (e.g., agriculture, water, etc.) are discarded. These categories represent between 50% and 65% from the total number of retrieved categories in each collection.