ADVANCED CONCEPTS IN POLARIMETRY – PART 2 (Polarimetric Target Classification)

There is currently widespread interest in the development of radar sensors for the detection of surface and buried targets and the remote sensing of land, sea and ice surfaces. An important feature of electromagnetic radiation is its state of polarisation and a wide range of classification algorithms and inversion techniques have recently been developed based on the transformation of polarisation state by scattering objects. There are three primary ways in which multi-parameter radar measurements can be made: multi-frequency, single or multi-baseline interferometry and multi-polarization. While several airborne systems can now provide diversity over all three of these, it is the combination of polarimetry with interferometry at a single wavelength that forms the central focus of future challenges in developing new and original data processing. The main reason for this is the imminent launch of a series of advanced satellite radar systems such as PALSAR, an L-band SAR sensor on board the NASDA ALOS satellite and Radarsat II, a C-band polarimetric sensor. These are typical of a new generation of radars with the potential for providing data from various combinations of polarimetry and interferometry. This paper seeks to review recent progress in polarimetric and interferometric SAR data processing, covering advances and addressing the important topic of classification of polarimetric SAR data. Indeed, classification of Earth terrain components within a full polarimetric SAR image is one of the most important applications of Radar Polarimetry in Remote Sensing. However, the selection of radar frequency and polarization are two of the most important parameters in synthetic aperture radar (SAR) mission design. For a particular application, it is desirable to optimally select the frequency and combination of linear polarization channels, if a fully polarimetric SAR system is not possible, and to find out the expected loss in classification and geophysical parameter accuracy. In the first part of this tutorial, we quantitatively compare classification accuracies between fully polarimetric SAR data and partial polarimetric SAR data, for P-, Land C-band frequencies. Additionally, to understand the importance of phase differences between polarizations, we compare the correct classification rates using the complex channels versus intensities channels. The second part of this lecture is dedicated to the presentation of different unsupervised classification methods that have been proposed during the last decade, based around the combination of the H / A / α Polarimetric Decomposition Theorem (S.R. Cloude and E. Pottier, 1997) and the maximum likelihood classifier based on the complex Wishart distribution for the covariance matrix (J.S. Lee et al., 1994). Unlike this approach classifies pixels statistically and ignores their scattering characteristics, a new segmentation that has better stability in convergence and preserves the homogeneous scattering mechanisms of each class and the purity of dominant polarimetric scattering properties for all pixels in a class will be presented and discussed (J.S. Lee et al., 2003). This algorithm uses a combination of a scattering model based decomposition developed by Freeman and Durden and the maximum likelihood classifier based on the complex Wishart distribution. Finally, an unsupervised classification process, gathering polarimetric and interferometric SAR data is presented. Data acquired in polarimetric and interferometric modes have complementary characteristics; their joint use in a classification process provides significantly higher performance and the resulting images show significant improvements compared to the strictly polarimetric case (L. Ferro Famil et al., 2001).