Volume II-7
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., II-7, 55-60, 2014
https://doi.org/10.5194/isprsannals-II-7-55-2014
© Author(s) 2014. This work is distributed under
the Creative Commons Attribution 3.0 License.
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., II-7, 55-60, 2014
https://doi.org/10.5194/isprsannals-II-7-55-2014
© Author(s) 2014. This work is distributed under
the Creative Commons Attribution 3.0 License.

  19 Sep 2014

19 Sep 2014

Classification of Urban Settlements Types based on space-borne SAR datasets

T. Novack and U. Stilla T. Novack and U. Stilla
  • Technische Universitaet Muenchen, Faculty of Civil, Geo and Environmental Engineering, 80333 Munich, Germany

Keywords: Urban structures types, Synthetic Aperture Radar, Feature selection, Histogram of Oriented Gradients

Abstract. In this work we focused on the classification of Urban Settlement Types (USTs) based on two datasets from the TerraSAR-X satellite acquired at ascending and descending look directions. These data sets comprise the intensity, amplitude and coherence images as well as binary images representing man-made structures. In accordance to most official UST maps, the urban blocks of our study site were considered as the analysis units. The urban blocks were classified into Vegetated Areas, Single-Family Houses and Commercial and Residential Buildings. As image attributes, Histogram of Oriented Gradients (HOGs) calculated in nine angles out of each image of our dataset were used. The descriptive statistics of the geometrical features of man-made structures inside the blocks were also used as image attributes. The pertinence of HOGs features for the concerning classification task is corroborated by feature selection algorithms used in the quantitative exploratory analysis of the attributes. 76 % of 1926 blocks were correctly classified and a Kappa index of 0.60 was obtained. As in some of the urban blocks from our study site more than one UST class are found, we propose a method for detecting blocks with mixed classes and segmenting them into more objects whose classification becomes hopefully more stable and accurate.