Feature Extraction from Hyperspectral Data Using ICA
209
H yperspectral Data
x : n bands
+
Principal
Component
Analysis
•
I
x' : n uncorrelated bands
I
1
Undercomplete
Independent
Component Analysis
+
I u : m independent components I
Fig. 8.3. Undercomplete independent component analysis based feature extraction (UrCAFE) algorithm
PCA
PCA
a
Band Drop
b
UICA
ICA
Fig. 8.4a-c. Schematic description of the two rCA based feature extraction algorithms along
with the original PCA algorithm. a PCA, b rCA-FE, c urCA-FE
ofUICA-FE (Fig. 8Ac), the band elimination step is replaced by the direct lCA
computation. Even though this step is time consuming, it has the potential of
extracting more informative features than PCA for further processing.