|
|
|
|
Ë
|
i
1
|
Fig. 4 Rollei 6006 metric
Due to the fact that there was no a-priori knowl-
edge about the spectral behavior of the different
materials, no special part but a wider range of
the spectrum was selected.
F C4
20
1
de
S00 700 900 2 [nml
Fig. 6 Transmission curvature of MKF6 filters
(from [Kronberg 1985])
In general, metal interference filters taken from
the multispectral camera MKF6 of Carl Zeiss Jena,
Fig.5 Rollei RS1 Réseau-Scanner
which mainly had been developed for remote
sensing in geology, were used, replenished by
some ORWO gelatine filters. Fig. 6 shows the
transmission curvatures of the used MKF6 filters.
It was tried to record images in the UV- region
(315nm - 380nm), the visible spectrum (380nm -
780nm) and the near infrared (680nm - 1200nm).
Given that the CCD-sensors are less sensitive in
the UV-region, with the RSC and the CS1, only the
acquisition of images in the visible and in the
near infrared was possible. By using different
films for the Rolleiflex 6006 metric, it was pos-
sible to obtain images that covered the whole
spectrum. Altogether, 11 different filters were
used with the different cameras.
During the survey, favourable constant diffuse
illumination with little shadow prevailed.
Recording | Area Image Format Pixelsize Spectral
System [pixel] image object Range
[um] | [mm] [nm]
RSC 1 2048 x 2048 15 6 400-1150
1 512 x 512 80 24 ORWO VF-45
6006/ 1024 x 1024 40 12 330- 730
RS1 2 510 x 512 80 55-80 |ORWO I-750
1024 x 1024 40 20-30 680- 800
CS1 1 768 x 512 8 21 400-1150
2 768 x 512 8 26-36
Table 1: Image Acquisition Conditions
3.3 Image evaluation
The multispectral image evaluation was effected
with the software package ERDAS(©), which was
originally developed for the evaluation of remote
sensing data. The software is implemented at a
SUN-workstation of the IPF at the Technical
University Dresden.
In a first step all corresponding images were
geometrically transformed to a common basis. In
order to reduce the amount of data, principal
component transformation was effected with the 7
- 8 original bands; for further evaluation, the
first four synthetic channels were used.
The spectral behaviour of the observed material
was unknown and a determination of training-
areas for a supervised classification became
difficult. Therefore, for a first analyse a
minimum distance operator was used for an
unsupervised classification of the images.