International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
rest are MNFs for other classes. Two MNFs with high
variances in oil spill are required to process the PU method to
be used in a 2-D scatter plot. Eigenvalue weighting is used to
compare the average spectrum of oil in the raw data and the
MNF spectra of oil. Both spectra are selected from the same
pixel location. The statistics provide clues to spectral features
contributing to the classification. The oil slick spectra and the
associated spectra in the image are examined for weighting
using the high variance (eigenvector) portion of the spectra.
2.3.2 Data Projection
Background
^
=
pus
ha
en
5
=
>
#3
e
|
[st]
js
De]
D
[zn
prem
Ea
©
jw}
©
a
=
D]
ue
E
h
Oil/Water
Band 53
23 pla br 5 28
N CES
Figure 9 (a) PU Results; Radiance Pixels Projection
In Figure 9 (a), the radiance (raw) data projection is for band 9
(visible) in the X axis and band 63 (near IR) in the Y axis.
These two bands were selected with minimum correlation to
maximize the variance between targets. It shows a good
separation in the background cluster (mountains) appearing in
brown and very poor separation between the oil and water
clusters appearing in the lower corner in blue, green, and
yellow.
78
Band 12 (fODOBOE01 p03_r02 =c03.1.imgkaanta_s
DP PT LT Te pe
—28,94 —5 558 9 e c IZ
MNF ERG ORI TN NE ET Amq):santa. se
Figure 9. (b) PU Results; MNFs Pixels Projection.
In Figure 9 (b), the vertical axis corresponds to the separation
between the targets (green = oiled water, red = thick oil slicks,
yellow = polluted water and streaks) and the background (white
cluster), while the horizontal axis maps inter-target separation.
The plot shows the optimal projection selected of the data using
2 MNF bands, where the background composite ROIs is
centered at the 0, 0 mean.
bj sx
Figure 10.The PU Results; The Abundance of Oil Materials
The agreement between the observed data and the 2-D scatter
plot indicates the successful derivation of the optimal projection
that the targets are optimally separated, and the multi-
component background is fully compressed. Spectral
signatures for oil spill types can be generated from the 2-D
scatter plot clusters.
Figure
11 A3
Bands Composite Image for a Complex Stage of the Oil Spill
1331
