clas-
Jera-
es is
each
lickly
Je in
s not
inder
s will
e eli-
Yodel
ope-
1 and
ation
figu-
1bset
ondi-
sting
al oc-
ilrea-
ints,
5. DISCUSSION
One of the most striking observations is the texture
dependency of the class finder. At a given reliability
level the Landsat TM image data of the area under
research will predominantly produce classes for the
sandy and ‘fresh’ eroded material. Because of the low
frequent changes in the spectral properties which are
not disturbed to such an extent by the relief like the
hardrocks, the primary aim in this research.
The computation time is also an important factor
worth to consider if this algorithm is to be applied.
One should be aware of the requirements for the sy-
stem, especially when used to compute large pixel
arrays.
6. CONCLUSION
The classification of rockunits in an arid area with
rapid change in relief and lithology can not be maintai-
ned by the applied procedure alone. There has to be
an accompanying image processing scheme to sup-
press the influence of texture and to prevent a rather
frustrating result in the way of non-classified hard-rock
units. An unsatisfying imagination for almost every
geologist is mapping in such a rough terrain structure.
The program is sensitive to the coefficients of varian-
ce and correlation and can be optimized by several
passes through the image data. Still a little bit of inter-
pretation work has to be conducted to select the best
class-site determination with reference to the classifi-
cation problem at hand. On the other hand it is impor-
tant to realise that the imaging conditions, to which
one has to be aware are very unfavourable because
sand often covers the rockunits in thin layers and the
influence of long term regional polymetamorphism
gives problems difficult to be solved especially if Fe-
ore minerals are finely distributed. But nevertheless
the algorithm seems to be a very reasonable tool for
the classification of soils with smooth transitions into
each other and low texture frequencies or on data
which have been relief corrected with less impact on
texture patterns.
The primary advantage of this scheme is the possibili-
ty to use unlimited number of spectral bands to get
classes and the possibility to influence variance, stan-
dard deviation and the degree of correlation of the
spectral ‘pixel vector’ in its neighborhood.
For further developments it is possible to obtain the
C-source code of the program.
Acknowledgments
This research was funded by the Deutsche
Forschungsgemeinschaft (German Research Foun-
47
dation) in the Special Research Project 69 (SFB 69).
Many thanks to Prof. Dr. F.K. List for supervising this
research and to our colleague R. Schóle for mainten-
ance of the computer system and support in the pro-
gramming task.
Address of Authors
FU Berlin, Geoinformatik, Malteserstr. 74-100, D -
12249 Berlin.
Tel.: +49 (0)30 7792 578, Fax: +49 (0)30 7752 075.
Email: 0308338578-1@t-online.de
REFERENCES
BORTZ, J. 1989: Statistik für Sozialwissenschaftler.
790p.. Berlin (Springer).
HAENISCH, H., KENEA, N.H. & OTT, N. 1996: Tectonic
Development of the Southern Red Sea Hills of Sudan
- Evidence from Landsat TM-Mosaic Interpretation. In:
WG VII/4, this volume.
MCCAFFREY, T.M. & FRANKLIN, S.E. 1993: Automated
Training Site Selection for Large-Area Remote-Sen-
sing Image Analysis. Computers & Geosciences, Vol.
19, No. 10, pp. 1413-1428.
PITAS, l. 1993: Digital Image Processing Algorithms.
362 p.. Heamstead (Prentice Hall International (UK)
Ltd).
SCHULZ, B.S. 1990: Analyse der statistischen Voraus-
setzungen zur Klassifizierung multispektraler Daten.
ZPF, 3, pp. 66-74.
SCHULZ, B.S. 1992: Fully Automated, High-Resolution
Classification of Remotely Sensed Digital Multispec-
tral Recordings. In: International Archives of Photo-
grammetry and Remote Sensing, Washington, U.S.A.,
Vol., pp. 912-913.
SCHULZ, B.S. & WENDE, C. 1994: Vollautomatische,
hochdifferenzierende, pixelweise Klassifizierung multi-
spektraler Bilder, Konzept, Methode, Ergebnisse. In:
F.K. List (ed.) 13. Wiss.-Techn. Jahrestagung der
DGPF 1993, S. 193-198. Berlin (DGPF).
WOLF, H. 1968: Ausgleichsrechnung und die Methode
der kleinsten Quadrate. pp. 509-523. Bonn (Ferd.
Dümmler Verl.).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
