Support Vector Machines for Classificationof Multi- and Hyperspectral Data
253
Radial Basis Function Kernel
100
90
80
>70
u
e ::I 60
u
u
50
40
.. 30
>
0
20
~
/
y:r
d
Y
~
d
~
~
~
~
~
10
Penalty Value (C)
Fig.IO.1S. SVM performance using the RBF kernel applied on the hyperspectral image
Sigmoid Kernel
100
90
80
>u
70
e ::I 60
u
u
50
«
E
40
II)
30
>
0
20
h
r
~
d
\
y:f
\
L
\
/
'\
~
\
~
"\
~
10
0
Penalty Parameter
Fig. 10.19. SVM performance using the sigmoid kernel applied on the hyperspectral image
are preferred to perform efficient classification. This is particularly important
in case of high dimensional data such as hyperspectral remote sensing images.
Thus, for the hyperspectral image, the SVM with the RBF and polynomial
degree 2 kernels performed the best with an accuracy of 97%. Other kernels
performed their best with an accuracy ranging from 70% to 96%. These results
illustrate that the choice of kernel function affects the performance of SVM
based classification.
253
Radial Basis Function Kernel
100
90
80
>70
u
e ::I 60
u
u
50
40
.. 30
>
0
20
~
/
y:r
d
Y
~
d
~
~
~
~
~
10
Penalty Value (C)
Fig.IO.1S. SVM performance using the RBF kernel applied on the hyperspectral image
Sigmoid Kernel
100
90
80
>u
70
e ::I 60
u
u
50
«
E
40
II)
30
>
0
20
h
r
~
d
\
y:f
\
L
\
/
'\
~
\
~
"\
~
10
0
Penalty Parameter
Fig. 10.19. SVM performance using the sigmoid kernel applied on the hyperspectral image
are preferred to perform efficient classification. This is particularly important
in case of high dimensional data such as hyperspectral remote sensing images.
Thus, for the hyperspectral image, the SVM with the RBF and polynomial
degree 2 kernels performed the best with an accuracy of 97%. Other kernels
performed their best with an accuracy ranging from 70% to 96%. These results
illustrate that the choice of kernel function affects the performance of SVM
based classification.