A Self-Initializing PolInSAR Classifier Using Interferometric Phase Differences

This paper describes an unsupervised classifier for polarimetric interferometric synthetic aperture radar (PolInSAR) data. Expectation maximization is used to estimate class parameters that maximize the likelihood of observations in an input data set for a given number of classes. Polarimetric information, in the form of coherency matrices, and interferometric information, in the form of complex coherences, are taken into account. Differences in interferometric phase across different polarization states are explicitly modeled to make the classifier sensitive to the vertical structure of the scene under observation, and the distribution over such phase differences is introduced. The classifier is self-initializing, in that it does not rely on decompositions or thresholds. Classification results obtained for real polarimetric interferometric data are presented and discussed.