SEMI-AUTOMATED CLASSIFICATION OF URBAN AREAS BY MEANS 
OF HIGH RESOLUTION RADAR DATA 
T. Esch, A. Roth 
German Aerospace Center DLR, German Remote Sensing Data Center DFD, 82234 Wessling, Germany- 
(Thomas.Esch, Achim.Roth)@dlr.de 
Commission VI, WG VI/4 
KEY WORDS: High Resolution SAR, Urban Areas, Detection, Contextual Analysis, Texture, Automation 
ABSTRACT: 
Almost two thirds of the world’s population will live in cities by 2030. Thus, human settlements typify the most dynamic regions on 
earth. To cope with this development urban planning and management requires up-to-date information about the various processes 
taking place within the urbanised zones. As radar remote sensing allows for the collection of areal data under almost all weather and 
environmental conditions it is predestined for a frequent and near-term retrieval of geo-information. 
In view of the future TerraSAR-X mission the German Remote Sensing Data Center (DFD) researches into the potential use of high 
resolution radar imagery. In this context our main objective is to develop concepts for a mostly automated detection and analysis of 
human settlements by means of high resolution SAR data. The studies are based on multi-frequency single-, dual- and quad- 
polarised SAR data recorded by the airborne Experimental Synthetic Aperture Radar (E-SAR) system of the German Aerospace 
Center (DLR). This paper presents first results of an object-oriented approach towards a semi-automated identification of built-up 
areas based on single-polarised E-SAR X-band imagery. The basic concept includes an optimised speckle suppression procedure in 
order to improve and stabilise subsequent image analysis as well as the development of an object-oriented classification scheme for 
an automated detection of built-up areas from high resolution X-band imagery. 
The developed concept for a semi-automated extraction of urban areas from single polarised X-band data yields promising results. 
Built-up areas could be detected with an accuracy of 86%, 85% and 91% for three flight tracks. While the main body of the 
settlements could be identified with an accuracy of more than 9096 inaccuracies were mainly associated with flanking parks, 
recreation areas and allotments. 
1. INTRODUCTION (Henderson, Xia, 1998; Kressler, Steinnocher, 2001; Ryherd, 
Woodcock, 1996). While most of these analyses were based on 
Numerous studies have focused on the utilisation of satellite conventional pixel-based techniques recent studies have 
imagery to analyse human settlements, monitor urban sprawl or increasingly focused on object-oriented approaches (De Kok, 
map urban land use patterns and infrastructure. Most of these Wever, Fockelmann, 2003: Kressler, Steinnocher, Kim, 2002: 
applications deal with an analysis of regional or local Hofman, 2001). These more sophisticated techniques provide 
phenomena based on spectral characteristics or indices derived possibilities to describe and utilise the geometric, textural and 
from high resolution optical satellite imagery (Ehrlich, Lavalle, especially contextual properties of the real-world objects in the 
Schillinger, 1999; Masek, Lindsay, Goward, 2000; Ridd, Liu, classification process. 
1998). In the context of urban applications radar imagery has 
played a minor role since the strong dependence of radar With the perspective of the future TerraSAR-X satellite (Roth, 
reflectance on geometrical characteristics of the illuminated 2003) this study aims at the development of a concept for the 
scene has hampered its interpretation significantly. In radar automated extraction of built-up areas based on high resolution, 
images the agglomeration of di- and trihedral corner reflectors single polarised X-band imagery. The main emphasis is placed 
in urban environments makes these regions standing out as on both, the improvement of image pre-processing with respect 
clusters of more or less bright signal returns. This effect has to a subsequent object oriented image analysis and the 
been used for monitoring the urban footprint and subsequently development of a robust classification scheme for a mostly 
estimating socio-economic characteristics (Haack, 1984; automated identification of human settlements. In the first part 
Henderson, Xia, 1998). In this regard polarised and/or multi- of this paper an overview on the methodological concept is 
frequency radar data have proven to be particularly valuable. given. Then preliminary results of the developed approach are 
Radar imagery has also been used successfully in combination shown. Finally an outlook on the further research is given. 
with optical data (Forster, Ticehurst, 1994; Weydahl, Becquey, 
Tollefsen, 1995). 
The strong dependence of the signal return on the geometrical 2. METHODOLOGY 
properties of the illuminated objects results in an ambiguity of 
its intensity signature with respect to its physical characteristics. This section provides a description of the method to detect 
This demands the use of alternative features for a distinct built-up areas on the basis of high resolution single-polarised X- 
classification. band imagery automatically. 
The applicability of textural information in urban applications 
has already been shown for optical and radar based imagery 
478 
Interna 
2.1 B: 
The nx 
e th 
su 
la 
e th 
Subseq 
optimis 
Along 
used f 
itself 1 
origina 
classifi 
Image 
This te 
spectra 
form in 
analysi 
applica 
It facil 
their fu 
charact 
area si 
oriente 
single | 
the init 
speckle 
from an 
Subseq 
created. 
identifi 
classifi 
spectral 
hierarcl 
neighbc 
image 1 
nearest 
perform 
protoco 
automa 
2.2 Te 
The X-1 
of Ludy 
Experin 
DLR (N 
the Rh 
largest « 
During 
polarise 
recorded 
of 20° - 
resolutk 
of urbar 
In order 
biodivei 
a data 
region. 
photogr