Classification of polarimetric synthetic aperture radar images using fuzzy clustering

Abstract

Clustering is a well known technique for classification in polarimetric synthetic aperture radar (POLSAR) images. Pixels are represented as complex covariance matrices, which demand dissimilarity measures that can capture the phase relationships between the polar components of the returns. Four dissimilarity measures are compared to judge their efficacy to separate complex covariances within the fuzzy clustering process. When these four measures are used to classify, a POLSAR image, the measures that are based upon the Wishart distribution outperform the standard metrics because they better represent the total information contained in the polarimetric data. The Expectation Maximization (EM) algorithm is applied to a mixture of complex Wishart distributions to classify the image. Its performance matches the FCM clustering results yielding a tentative conclusion that the Wishart distribution model is more important than the clustering mechanism itself.