in case of an unsupervised classification, define a
number of classes which will be found after a certain
number of pixels have been processed (e.g. cluster
analyses). They do not account for the effect of the
image content that may leave part of the image un-
processed, at the later stages. This is one of the ai-
med improvements of the implicated algorithms.
Originally the application under concern has been
choosen for the evaluation of a terrain and atmosphe-
ric influences suppressing image processing scheme.
Eph Zo son des IE A y
t emu vat E» á
TY OES NT oe ….
eR > RS :
XN M
NS
Re
B :
Metabas03
fai Gneiss-granitic rocks
3 Metamafitic to ultramafitic rocks
Ophiolitic, metagabbroic rocks
Talcschists
[7] Schistose metamatitic rocks
E] Dikes
Haraid Haenisch 1995
[1 Wadi deposits N
Quatenary undifferentiated A
! Metagabbro
| F3 Metagabbro, partly covered by debris
|
fig.1: Geological sketchmap of the area under considera-
tion.
This method is under development at the Free Uni-
versity of Berlin. The basic ideas have been proposed
by SCHULZ et al. (SCHULZ 1990, 1992 and SCHULZ &
WENDE 1993) as part of the program "Objektbezoge-
ne Informationsgewinnung aus Fernerkundungsdaten
zum Aufbau eines digitalen Landschaftsmodells" -
IFAG, Frankfurt/Main. The authors application has
been set to work at the Institute for Geoinformatics of
the Free University of Berlin. But during the implimen-
tation it appeared that there could be a possibility for
use as rock discrimination scheme under geological
aspects, a theme difficult to handle, if one wants to
44
see varieties of rock units and not only to distinguish
between rocks in general and unconsolidated sedi-.
ments.
2. DATA AND RESOURCES
The dataset to be classified is a subset of the Landsat
TM image 171 - 47 (25. Dec. 1988). The test site is
part of the southern Red Sea Hills of Sudan aproxi-
matly 150 km SW of Port Sudan.
The rocks which can be observed there are at most
intensely, alterated amphibole to greenschist facies
metabasitic units with some additional deeply weathe-
red syn- to late-tectonic migmatites and granites. The
meta-basitic rocks are talc schists, chlorite-serpentini-
te schists, chlorite schists, metabasalts and metag-
abbros to meta-ultramafitites (fig. 1).
The hilly country side is disected by wadi water run-
offs. It shows in addition to the overall variety of rocks
a high frequent change in lithology with irregular repe-
tition and thickness. This was another motivating
ground to apply an automatic classification scheme.
The situation has been formed by tectonic processes
beginning with late Proterozoic rifting and accretion
developments in northeastern Africa during the Pan-
African tectono-thermal phase (HAENISCH et. al. 1996,
this volume).
Implementation of the algorithm has been realised in
'C' on a Sun SparcStation 10 with 96 MB mainmemo-
ry. In order to get a simple format the dataset has
been transformed with Erdas Imagine 8.2 in a generic
binary format with band interleaf by line mode (BIL), a
format almost equal to the Erdas 7.4 LAN-File repre-
sentation without the 128 byte header. The data con-
version was carried out with im-/export-modul of E-
rdas Imagine 8.2.
To have an easy access to problems imerging during
the development, the realisation has been constituted
of three parts. Calcta is the program part to find the
training areas, ca/ccías develops the classes as des-
cribed below, and calcres attaches the pixels to the
given classes.
3. CONCEPT
In this particular approach the classification of multi-
spectral image data is devided into three main
functions 'search for the smallest homogeneous set of
training areas', reduction of the set of smallest homo-
geneous training areas to classes', and 'attaching the
image pixel values to the related classes (fig.2).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
— gm Ar 1]
—— 0 IN a AN a MA) pede
— A ga
FX FX "rT AA s (nA
a—- emp N NOON BÀ may meme AP
-— |. Seu a JA
