Polarimetric SAR image classification with the maximum curvature of the trajectory in the eigen space converted from the polarization signature K. Arai * , J. Wang Saga University, 1 Honjo, Saga 840-8502, Japan Received 28 July 2004; received in revised form 16 November 2005; accepted 15 February 2006 Abstract A method for polarimetric SAR image classification based on maximum likelihood classification with the maximum curvature of the trajectory of in the eigen space converted from the polarization signature is proposed. One of the examples of the sea ice classification is shown. The classification performance of the proposed method is compared to the existing methods based on the maximum likelihood classification with (1) fully polarimetric SAR data (received power), (2) the received + H + A + (a), (3) the received power + odd + even + diffuse of the scattering components. The RI-SAR data of the sea of Okhotsk and Tsukuba in Japan which are acquired on 23 February 1999 and on September 30 1997 are used. It is found that the proposed method is superior to the existing methods in terms of the classification performance at the 95% of confidence interval as well as the kappa coefficient. Ó 2006 COSPAR. Published by Elsevier Ltd. All rights reserved. Keywords: Polarimetric SAR; Image classification; Sea ice; Polarization signature; Maximum likelihood classification 1. Introduction Radar polarimetry is useful for characterizing dielectric constant, slope/aspect of the surface, directionality of arti- ficial objects and so on through clarification of the interac- tion mechanism between electromagnetic (EM) wave and the targets (Ulaby and Elachi, 1990; Zebker and vanZyl, 1991). Polarimetric SAR image classifications with the fol- lowing three components of the fully polarimetric SAR data were proposed (Mott, 1992; Henderson and Lewis, 1998). 1. HH (transmit EM wave with horizontal polarization (H- polarization) and receive EM echo with H-polarization). 2. HV (transmit EM wave with H-polarization and receive EM echo with vertical polarization (V-polarization)). 3. W (transmit EM wave with V-polarization and receive EM echo with V-polarization). On the other hand, the extraction methods of the scat- tering characteristics by applying eigen decomposition to the covariance matrix derived from the scattering matrix which is calculated from the aforementioned three compo- nents were proposed (Krogager, 1993). Furthermore, the classification methods with single/double/multiple, odd/ even/diffuse, and odd/even/Bragg/multiple scattering com- ponents were proposed (Dong, 1987) while the classifica- tion methods with sphere/di-plane/helix, and sphere/ Bragg/double of scattering components which are based on spherical polarization were proposed (Zebker and van- Zyl, 1991; Krogager and Czyz, 1995). The aforementioned methods were reviewed (Cloude and Pottier, 1995). More- over, the classification method with the entropy (H) which is defined with the sum of the first to the third eigen values and the ratio of each eigen values, the anisotropy (A) which is defined as the ratio of sum and subtraction of the second 0273-1177/$30 Ó 2006 COSPAR. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.asr.2006.02.029 * Corresponding author. E-mail address: arai@is.saga-u.ac.jp (K. Arai). www.elsevier.com/locate/asr Advances in Space Research 39 (2007) 149–154