AN ALTERNATIVE FOR ENTROPY-ALPHA CLASSIFICATION FOR POLARIMETRIC SAR IMAGE

摘要

In this work we discuss SAR target entropy and alpha angle relations to other scattering covariance matrixrncharacteristics and similarity invariants. It is shown that the sum of squared elements of the coherency matrix,rnnormalized by its trace and determinant, has many common features with target entropy parameter. The first element ofrnthe matrix is very similar to alpha angle parameter describing scattering mechanism. Possibilities to use the sum ofrnsquared elements, determinant and first element of normalized coherency matrix for classification are studied. It appearsrnthat classification schemes very similar to entropy-alpha can be established. However, classification results differrnslightly from those of entropy-alpha classification as here discussed two-parameter classifications depend on threernvariables, although parameters are in all cases the same. As an example, NASA/JPL AIRSAR L-Band image of the SanrnFrancisco Bay was classified with both proposed schemes and original entropy-alpha classification. The size of the usedrnimage was 224 x 256 pixels. The new algorithms classified 97% and 96%, respectively, of pixels to the same classes asrnentropy-alpha classification. The discussed similarity invariants are straightforward to calculate and they have beenrnused to describe covariance matrix properties in statistics. Virtually are proposed classification algorithms equivalentrnwith entropy-alpha classification because all three use the same amount of information from covariance matrix.rnHowever, proposed parameter pairs are much easier to calculate, as they do not require the computation of eigenvaluesrnand eigenvectors.