An Unsupervised Segmentation With an Adaptive Number of Clusters Using the $SPAN/H/\alpha/A$ Space and the Complex Wishart Clustering for Fully Polarimetric SAR Data Analysis

In this paper, an unsupervised segmentation is proposed for fully polarimetric synthetic aperture radar (SAR) data analysis. The backscattering power SPAN combined with H/alpha/A is used to obtain the initial cluster centers. We use the Wishart test statistic to perform an agglomerative hierarchical clustering to obtain the segmentation results with different numbers of clusters. The appropriate number of clusters is automatically estimated using the data log-likelihood (Lm), and the resulting images with the estimated number of clusters are the final segmentation results. The experiments show that the SPAN has additional information that is not contained in H/alpha/A, and this information could be useful for the initialization. The number of clusters seems to be a crucial point for the segmentation, which will affect the segmentation performance. It is also shown that the data log-likelihood has the potential ability to reveal the inner structure of fully polarimetric SAR data.