4. THE APPLIED SOFTWARES
I used the DigiTerra IMAGE software for the orientation,
rectification and classification purposes. It was developed by my
college, Kornél Czimber and the DigiTerra Engineering Office,
and has its own graphic user interface, and the most common
functions of image processing. For the GIS analysis and layouts
the ESRI-Autodesk ArcCAD v11.3 and the ESRI ArcView 3.0
software was used..
5. THE ORIENTATIONS
The Landsat TM image was georeferenced, by the FÓMI RSC.
The SPOT image was not oriented. | made the orientation with
the DigiTerra-IMAGE software. I pick the ground control points
(GCPs) from the DFBM. The accuracy of the orientation was
0.38 pixel (3.8 m). After the orientation a bilinear resampling
with 10 m ground resolution was made. Then the Brovey
transform was prepared with the Landsat TM and SPOT Pan.
According to certain opinion (Koch et al, 1997) the IHS
transformation gives better results, but - as it appears below - I
preferred the Brovey transformation. The result, a 10 m ground
resolution 7 band multispectral image is very nice for visual
interpretation, but hardly usable for classification methods.
I produced ortophotographs from the aerial photographs. Instead
of using stereo pairs, I applied single photographs with digital
elevation model (DDM-10). The orientation points were from the
DFBM, too. The orientation parameters of the P1 are showed in
Table 2.
A= 0.27187 °
B= -3.78701 °
r= 0.757248
focal = 152.7 mm
of aerial Z= 4842.4 m
Table 2: The orientation parameters of the P1 photo.
The average accuracy of the orientation was 2.08 pixel which
means 5.55 m. After the orientation I made a bilinear resampling
with 3 m ground resolution.
The orientation parameters of the P2 are showed in Table 3.
-0.983063
-1.70205 ?
3.41245 ?
152.13 mm
Z- 1257.37 m
Table 3: The orientation parameters of the P2 photo.
The average accuracy of the orientation was 13.9 pixel which
means 9.73 m. It doesn't seem so good, but according to the
accuracy of the DFBM it isn't bad. The other reason is the
resolution used, 300 dpi at the scanning, but our hardware
possibilities are limited. After the orientation I made a bilinear
resampling with 0.5 m ground resolution.
6. THE CLASSIFICATION METHOD
6.1 The developing of classes
I used the capabilities of the DigiTerra-IMAGE software in the
classifications. I made the forests classes based on the National
Forest Database, and the DFBM. I developed the polygons of
different classes from forestry point of view. Then I made buffer-
zones from the polygons with negative values (decreasing the
area, Figure 4.) for omitting the mixed pixels at the stand's
boundary.
Figure 4: Buffering polygons.
Then I put the emphasis on the refinement of the training areas.
The reason for this modification is the different updating time of
the DFBM and the National Forest Database and the different
creation time of the images. 10 forest classes were shaped at
first, these were the followings: robinia, sessile oak, hairy oak,
hornbeam, birch, beech, larch, scotch pine, austrian pine, spruce.
6.2 Classification of Landsat TM
The first step was the examination of the bands. The first band
was very noisy, the second band was striped, and because of the
georeference the strips were diagonal. | couldn't remove the
diagonal strips, becausc only ‘horizontal’ and ‘vertical’ strips are
removable by the software I used. So the Band I., II., and the VI.
- because of the insufficient resolution - were omitted from the
classification method. The next measures was the topographic
normalisation. The correlation between the bands and the DTM
was studied at first. | chose an area with homogenous vegetation
for the statistics. The results are summarised in Table 5.
The number | The inclination of the | The correlation
of the band regression line (m) coefficient
III. 0.03 0.51
IV. 0.46 0.87
V. 0.17 0.85
VII. 0.07 0.76
Table 5: The regression parameters between the bands and the
DTM.
It is apparent that the band IV. has the most correlation with the
DTM. An approaching DTM extraction is possible, when the
vegetation is homogenous or a land cover map is available.
366 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
Thei
para
I sta
class
algo
Thu:
horn
error
meth
shov
Tat
Ther
do nc
