ul 2004
l-scale
> third,
omplete
? image
assified
se" À
robust
basing
'eatures
coarse
1 of the
Ires, in
ver the
ullt-up
fields.
form in
fled to
, other
ned on
; to the
ouring
eristics
coarse
mation
ovided
: initial
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
4. RESULTS
The rule base developed on the basis of small subsets is applied
to each of the three flight track images to determine the
transferability of the classification scheme. The outcome is
validated by comparing the classification to the reference GIS
data set and the aerial photographs based on 500 randomly
distributed control points. Since the vector data isn't available
for all of the area covered, the accuracy for the respective
regions is determined solely by a comparison to the aerial
photographs.
The result shows that the built-up areas can be identified with
an overall accuracy of 86% for track 1, 85% for track 2 and
91% for track three. The classification for flight track 1 is
illustrated in Figure 3 along with the reference vector data on its
right. It can be seen that the main body of the settlements —
shown red in the reference map - is detected very accurately.
These areas could be classified with an accuracy of more than
90%. This promising result can be traced back to the fact that
these zones contain quite characteristic features like high
texture, significant spectral difference of the single structures to
their neighbours and the existence of multiple strong scatterers
and shadow areas. In this manner even recently built-up
development areas could be identified, which are not yet
included in the reference data (see blue arrows in Figure 3).
Errors occurred mainly in the context of highly structured areas
possessing diverse strong scatterers and significant shadow, e.g.
non-urban street crossings surrounded by groups of trees.
The most significant false assignments occurred in recreation
areas or allotments flanking the settlements. These zones -
represented by greenish colours in the reference map - could
only be identified when featuring significant texture along with
the existence of some bright scatterers. Those spots situated in
the far range region of the image lack these conditions due to
the decreasing influence of surface roughness in far range.
Consequently, these areas appear as dark, smooth zones without
significant scatters. Thus, they can not generally be separated
from areas in mid- and near-range featuring a significantly
lower meso-scale roughness in reality. In mid- and near range
the recreation areas and allotments possess the same
characteristics as smaller forest stands, groups of trees or rough
agricultural fields — consequently these zones remain
unclassified.
S. CONCLUSIONS
The study has shown the applicability of high resolution, single
polarised X-band imagery for the detection of built-up areas.
Nevertheless, the significant variation of radar backscatter
along the surface of a single structure (e.g. a building), the
interaction of the scattered waves with multiple objects, the
range-dependant effects of surface roughness and the visibility
of objects subject to the line of sight result in considerable
ambiguities of the recorded signature. Thus, the intensity
information has to be analysed in consideration of its spatial
context including the attributes of the surrounding objects as
well as the characteristics of the super ordinate structure instead
of solely regarding the backscatter signal. The object-oriented
approach has proven to be very efficient as it provides multiple
tools and features to address textural, contextual and
| hierarchical properties of image structures.
481
6. FUTURE PERSPECTIVES
A first future goal is the improvement of the presented
classification scheme in view of a more accurate detection of
recreation areas, parks and allotments. In addition the
development of comparable procedures for single-polarised C-
and L-band imagery is designated.
A second field of study will be dealing with the potential
improvements associated with an analysis based on a
combination of single-polarised X- and L-band data (multi-
frequency analysis). Finally the integration of polarimetric
decomposition properties considering the quad-polarised L-
band data is planned.
REFERENCES
Baatz, M., Schápe, A., 1999, Object-Oriented and Multi-Scale
Image Analysis in Semantic Networks. In: Proc. of the 2nd
International Symposium on Operationalization of Remote
Sensing, August 16-20, 1999. Enschede. ITC.
De Kok, R., Wever, T., Fockelmann, R., 2003. Analysis of
urban structure and development applying procedures for
automatic mapping of large area data. In: Carstens, J. (Ed.):
Remote Sensing of Urban Areas 2003, 41-46.
Ehrlich, D., Lavalle, C., Schillinger, S., 1999. Monitoring the
Evolution of Europe's Urban Landscapes. In: Proc. of the
IGARSS'99 in Hamburg. Vol. III, pp. 2705-2707.
Forster, B.C., Ticehurst, C., 1994. Urban Morphology
Measures from Optical and RADAR Remotely Sensed Data —
Some Preliminary Results. Proc. ISPRS Commission VII
Symp., Sept. 26-30 Rio de Janeiro, Brazil, Vol. 30, Part 7b, pp.
291-296.
Haack, B., 1984. L- and X-Band Like- and Cross-Polarized
Synthetic ^ Aperture Radar for Investigating ^ Urban
Environments. Photogram. Eng. & Remote Sensing, Vol.50,
No.10, pp. 1471-1477.
Haralick, R. M., 1979. Statistical and structural approaches to
texture. Proceedings IEEE, Vol. 67, no.5, pp. 786-803.
Henderson, F.M., Xia, Z.G., 1998. Radar Applications in
Urban Analysis, Settlement Detection and Population Analysis.
Principles and Applications of Imaging Radar (F.M. Henderson
and A.J. Lewis, eds.), Chapter 15. New York, pp. 733-768.
Hofman, P., 2001. Detecting urban features from IKONOS
data using an object oriented approach. In: Remote Sensing &
Photogrammetry Society (Ed.): Proceedings of the First Annual
Conference of the Remote Sensing & Remote Sensing Society,
28-33.
Kressler, F., Steinnocher, K., 2001. Monitoring | urban
development using satellite images. In: Jürgens, C. (Ed.).
Remote Sensing of Urban Areas. Regensburger Geographische
Schriften, Heft 35, 140-147.
Kressler, F., Steinnocher, K., Kim, Y., 2002. Urban land cover
mapping from Kompsat EOC panchromatic images using an
object-oriented classification approach. In: Proceedings of the
Third International Symposium Remote Sensing of Urban
Areas, Vol. 1, ISBN 975-567-219-X, pp. 219-226, Istanbul, 11-
13 June, 2002.
