Articles | Volume 3
Adv. Radio Sci., 3, 399–400, 2005
https://doi.org/10.5194/ars-3-399-2005
Adv. Radio Sci., 3, 399–400, 2005
https://doi.org/10.5194/ars-3-399-2005

  13 May 2005

13 May 2005

Principal Component Analysis In Radar Polarimetry

A. Danklmayer1, M. Chandra2, and E. Lüneburg3 A. Danklmayer et al.
  • 1Institut für Hochfrequenztechnik und Radartechnik, Deutsches Zentrum für Luft- und Raumfahrt, P.O. Box 1116, 82234 Weßling, Germany
  • 2Professur für Hochfrequenztechnik und Photonik, Technische Universität Chemnitz, Reichenhainer Str. 70, 09126 Chemnitz, Germany
  • 3Ernst Lüneburg, EML-Consultants, Georg-Schmied-Weg 04, 82234 Oberpfaffenhofen, Germany

Abstract. Second order moments of multivariate (often Gaussian) joint probability density functions can be described by the covariance or normalised correlation matrices or by the Kennaugh matrix (Kronecker matrix). In Radar Polarimetry the application of the covariance matrix is known as target decomposition theory, which is a special application of the extremely versatile Principle Component Analysis (PCA). The basic idea of PCA is to convert a data set, consisting of correlated random variables into a new set of uncorrelated variables and order the new variables according to the value of their variances. It is important to stress that uncorrelatedness does not necessarily mean independent which is used in the much stronger concept of Independent Component Analysis (ICA). Both concepts agree for multivariate Gaussian distribution functions, representing the most random and least structured distribution.

In this contribution, we propose a new approach in applying the concept of PCA to Radar Polarimetry. Therefore, new uncorrelated random variables will be introduced by means of linear transformations with well determined loading coefficients. This in turn, will allow the decomposition of the original random backscattering target variables into three point targets with new random uncorrelated variables whose variances agree with the eigenvalues of the covariance matrix. This allows a new interpretation of existing decomposition theorems.