Hyperspectral Classification Using ICA Based Mixture Model
Hyperspectral Data
~
~minate water absorption-i
~_ and noisy bands
~
I
Feature Extraction
PCA
SPCA
OSP
PP
Variance
V: .
Projection Projection
anance
order
order
Ranking Criterion
I I _ _ ~l ~~------,
[
Feature Selection
-------.------.-- ---.------.------------'
ICAMM ICAMM
and
and
K-means K-means
ICAMM ~CAMM
and
and
K-means K-means
[
223
Fig. 9.1. Proposed experimental methodology for unsupervised classification of hyper spectral data
9.3.1
Feature Extraction Techniques
Due to a large number of narrow spectral bands and their contiguous nature,
there is a significant degree of redundancy in the data acquired by hyperspectral sensors (Landgrebe 2002). In classification studies involving data of high
spectral dimension, it is desirable to select an optimum subset of bands, in
order to avoid the Hughes phenomenon (Hughes 1968) and parameter estimation problems due to interband correlation, as well as to reduce computational requirements (Landgrebe 2002; Shaw and Manolakis 2002; Tadjudin and
Landgrebe 2001; Richards and Jia 1999). This gives rise to the need to develop
algorithms for reducing the data volume significantly. From the perspective of
statistical pattern recognition, feature extraction refers to a process, whereby
a data space is transformed into a feature space, in which the original data