

EntropyAlpha classification alternative for polarimetric SAR image
Jaan Praks ^{(1)}and Martti Hallikainen^{(1)}
^{(1)}
Helsinki University of Technology,
P.O. Box 3000,
FIN02015 HUT,
Finland
Abstract
In this work we describe how coherency matrix simple
invariants can be used to create similar classification
parameter space to the wellknown EntropyAlpha
classification. Proposed parameter pair is very simple and
fast to calculate with almost any remote sensing oriented
image processing software if polarimetric data is given in
covariance matrix (or comparable) format. Calculations
employ only linear combinations of absolute values of the
coherency matrix elements. Proposed parameters are related
to entropy and alpha with some ambiguity and classification
results differ slightly from entropyalpha classification.
Classification results between entropyalpha and proposed
scheme differ only for 3% of an example image pixels. As an
example, NASA/JPL AIRSAR LBand image for the San Francisco
bay was classified with both algorithms. Size of the used
image was 224 x 256 pixels. Both algorithms classified 97 %
of pixels into corresponding classes. Highest
misclassification rates are for high entropy classes.
Method is suitable for using as fast approximation for the
entropyalpha classification or independent classification
scheme.
Full paper
