International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
conclusion of benefits of influence TIR mosaic (component)
on final data fusion. The combinations are: all images; all
images without TIR image; TIR and 3 VNIR components;
just 3 VNIR components; TIR, 3 VNIR and 2
Volterra/Unsharp components; 3 VNIR and 2
Volterra/Unsharp components; TIR, 3 VNIR and 2 Gram-
Schmid components; 3 VNIR and 2 Gram-Schmid
components; TIR, 3 VNIR and 2 Sobel components; 3 VNIR
and 2 Sobel components.
4.2. Class
| Figure 7. R channel from VNIR treated with
| Volterra/Unsharp filter
|
3.2. Principal components
From the above mentioned 11 images of the same area, 11
principal components are produced. In this way, another
information about specific area is produced, the first principal
component which, in this case, contains 82% of information
of the scene.
Figure 9. Ground thrut data
After sampling of scene on above mentioned combinations of
images of specific terrain and looking at the pictures taken
from the ground (ground truth data) the foloving conclusion
was made. We can recognize 5 classes on the scene and these
are: bushes and high grass; low grass; rocks, separatelly
stones and stone walls; bare land and snow. According to
sampling of scene on mentioned combinations, only the snow
can be well defined without any doubt in all combinations
with TIR component. For instance, in combination with only
3 VNIR components and combination with 3 VNIR and 2
Sobel components we can't do this. The second best class for
recognition is bushes and high gras. The recognition of this
class is also better in combinations with TIR component. The
deviation in recognition of this class is bigger in
combinations of images without TIR component. Shadows in
the hole in the ground (named Kapljuv), as well as below the
rocks, also present the big problem in determining the class
he because of the false information which we have got from this
of part of terrain. The lower influence of shadows on
; differentiation of class is in combination of images with TIR
ras Hu : i ;
| component. The best example for this is the bare land in
er. | K $
apljuv.
Iter pj
vi s: ; v s : s 4.3. Automatic classification
fter Figure 8. First principal components created of all layers
be | ; : Du
| © + The number of classes was defined and auto-classification
à | 4. CLASSIFICATION ie : ; sus
1 | with 6 class was done. The inputs for particular classification
MC | ; ; were combinations mentioned in chapter 4.1. with four
NIR | 4.1. Sampling of scene
adition combinations with foloving inputs: TIR, 3 VNIR, 2
Volterra/Unsharp i 2 Sobel components; 3 VNIR, 2
Volterra/Unsharp i 2 Sobel komponente; and second one:
TIR, 3 VNIR, 2 Gram-Schmid i TVI component; 3 VNIR, 2
Gram-Schmid i TVI components; 3 VNIR, 2 Gram-Schmid i
TVI components.
| From the above mentioned images of terrain the 10
combinations of images were done (5 with TIR mosaic and 5
without TIR mosaic). On this combinations of images, the
recognition of class was carried out. The combinations were
created with TIR and VNIR mosaic and their enhancements,
and the same combinations without TIR mosaic for getting
1257
