Feature Extraction from Hyperspectral Data Using ICA
211
a
c
d
g
j
Fig.8.6a-j. First 10 components with the highest variance data processed with peA
classes, it is unlikely that each of them will be represented by the orthogonal
data.
Next, we applied the urCA-FE algorithm. The resulting components are presented in Fig. 8.8. There is a considerable difference in class separation between
the rCA-FE and the urCA-FE results, with the latter producing components
that are almost identical to the result obtained when rCA was applied to the
full image cube (see Fig. 4.9). It is also interesting to note that urCA-FE is able
211
a
c
d
g
j
Fig.8.6a-j. First 10 components with the highest variance data processed with peA
classes, it is unlikely that each of them will be represented by the orthogonal
data.
Next, we applied the urCA-FE algorithm. The resulting components are presented in Fig. 8.8. There is a considerable difference in class separation between
the rCA-FE and the urCA-FE results, with the latter producing components
that are almost identical to the result obtained when rCA was applied to the
full image cube (see Fig. 4.9). It is also interesting to note that urCA-FE is able