250
10: Pakorn Watanachaturaporn, Manoj K. Arora
Polynomial Kernel of degrees 5 to 7
100
~ ~ -
90
o ~
=
l!r1!I'~
80
>[3/
(,l
70
f!
j
:l
60
(,l
(,l
50
<
~
40
-+-PoIy D5
5
30
-e-PoIyD6
20
-tr-Poly D7
10
0
.!'., #., <5>"
~.,
.,
., .,~ .,<5> .,# .,.!' ~
~.
~.
~.
~ '
~.
"
Penalty Value (C)
Fig. 10.12. SVM performance using the polynomial kernels of degrees 5 to 7 applied on the
multispectral image
Radial Basis Function Kernel
100
90
A
A
80 V' -v'
~
f!
70
:l
60
::I 50
<
~
40
~ 30
0
20
10
0
.!'., #., <5>"
~.,
., ., .,~ .,<5> .,# .,.!' ~
~ '
~ .
(>.
(>.
(>'
"
Penalty Value (C)
Fig. 10.13. SVM performance using the Radial Basis Function kernel applied on the multispectral image
the maximum. This shows that there is a limiting C value for all the kernel
functions.
In the case of hyperspectral data (see Fig. 10.15 to Fig. 10.19), generally,
a similar trend has been observed except for a couple of observations worth
mentioning. Although, all the kernels achieved a maximum accuracy of more
than 90%, this maximum occurs at different C values for each kernel. For