Hyperspectral Classification Using ICA Based Mixture Model
233
Table 9.5. Error matrix for the classification of data, preprocessed by OSP. a ICAMM classification algorithm and b K-means classification algorithm
a
ICAMM (OSP)
Reference map
Total
Cl
C2
C3
C4
C 1
430
48
59
184
721
Classification
C2
18
431
0
160
609
Map
C3
89
8
324
84
505
C4
57
153
0
1155
1365
Total
594
640
383
1583
3200
Producer's Accuracy (%)
72.4
67.3
84.6
73
User's Accuracy (%)
59.6
70.8
64.2
84.6
Overall Accuracy (%) = 73. 1
C 1 - Background, C 2 - Corn-notill, C 3 - Grass/Trees, C 4 - Soybeans-min
b
K-means (OSP)
Reference map
Total
Cl
C2
C3
C4
Cl
369
18
306
295
988
Classification
C2
3
14
0
89
106
Map
C3
51
0
76
4
131
C4
171
608
1
1195
1975
Total
594
640
383
1583
3200
Producer's Accuracy (%)
62.1
2.2
19.8
75.5
User's Accuracy (%)
37.4
13.2
58.0
60.5
Overall Accuracy (%) = 51. 7
C 1 - Background, C 2 - Corn-notill, C 3 - Grass/Trees, C 4 - Soybeans-min
we conclude that the AVIRIS data preprocessed by SPCA and classified by the
ICAMM algorithm, exhibits an improved overall classification performance.
Additionally, these results clearly show that the ICAMM algorithm outperforms the conventional K-means algorithm for the classification of AVIRIS
data considered in this experiment.
9.5
Summary
The primary issue that motivated this research is the limitation of Gaussian
mixture model based classification algorithms. The underlying Gaussian distribution assumption is limited in the sense that the Gaussian mixture model
exploits only second order statistics of the observed data to estimate the poste-