Before performing a new classification with these two new
bands their 2D Scatter Plots is calculated.
synthetic band 1
T DT T T T T
000 215 4M 6 46 8.62
synthetic band 2
Figure 11: 2D-scatter plot between the two synthetic
bands.
A new classification is done using the new bands:
- supervised classification;
- bands: channels 1, band1s and band2s;
- maximum likelihood method;
- . threshold: 2096
- classes: snow, green, vegetation, water, cultivated
field, rice-field, cloud.
Using channel 1 and the synthetic band1s and band2s the
classification does not get better because the behaviour of
snow-green mixed pixels become similar with the
behaviour of buildings and so the snow-green mixed
pixels are likely to be classified as building. Table 3 shows
building class behaviour:
Snow + green
Band Building ed adciguds
Band 1 136.7667 134.69
17.7053
channel; 1.4398 0.7507
channel; +0.3579
channel, 1.2829 1.010
channel; 40.2291
Table 3: comparison between the signatures of building
and the 50%snow-50%green mixed pixels in the channels
band1, band1s and band2s
As table 3 shows the half snow-covered and half green-
covered pixels are liable to be confused with building in
band, and bandas (ratio between channel 4 and channel
3). Only in bandys (ratio between channel 2 and channel
3) this mistake can be avoided because behaviours are
different.
7. SYNTHETIC BANDS REALISED WITH NON LINEAR
TRANSFORMATION
Other synthetic bands are calculated with non-linear
transformations in order to improve the bands choice. The
non linear transformations are very “tricky” because they
change the digital numbers distribution.
These new bands (bands, and banda,) are obtained by
squaring band? and band; respectively.
It is important to underline the channel 3 contribution for
snow identification, because in channel 3 snow and green
have quite different behaviours, the new values are then
divided by band; and so the new bands are the following:
2
bands, = (band?)
band;
2
bandas = (band4)"
band;
2D scatter plot between
band3s and band4s
0.00 2.96 593 8.8v 11.86
Fig. 12: 2D scatter plot between bandss and bands.
As figure 12 shows there is a low correlation degree
between these new synthetic bands and so they are
suitable to be used for a new classification using the
following parameters:
- supervised classification,
- . bands: channel 1, banda, and bandas;
- maximum likelihood method;
- . threshold: 2096
- classes: snow, green, vegetation, water, cultivated
field, rice-field, cloud.
The result of this new classification shows that using band
1, banda, and banda, the snow+green mixed pixels are
classified as rice-fields; and so one can see that even this
new bands combination does not improve the mixed
pixels classification too.
The following table contains the signatures of snow, green
and rice-field, in the new synthetic bands:
Band | Snow | Green | Rice- |snow green
field 2
180.1005 | 89.2856 | 121.8830 | 134.6930
Bandi | 430232 |+254259 | +5.5950
channel? | 06811 | 06704 | 0.8014 0.6757
channels | «0.2029 | «0.5086 | 10.6721
06326 | 06898 | 12111 0.6612
channel 42
channels +0.4032 | +0.4787 | +0.7987
Table 4: the spectral signatures using bandas and bandas.
332 Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
As 1
coni
The
the
con:
lake
sign
ovet
suci
8.N
Sinc
the
num
pos:
is fc
trea!
Mixe
whic
The
com
inact
very
havir
unce
Trad
abso
assic
pixel
only,
exist
the fi
This
assic
parti:
pixel:
ciass
repre
Fuzz
and
thing
beca
appre
uniqu
degre
are |
mem
The i
havin
pixels
assig
Figur
(first
consi
