Classification strategies for fully polarimetric SAR data of sea ice
University of British Columbia,
2356 Main Mall,
V6T 1Z4 Vancouver, BC,
(2) German Aerospace Center e.V., P.O. Box 1116, D-82230 Wessling, Germany
The potential of polarimetric SAR data for classification
purposes is widely recognised. More information on
scattering mechanisms is now available and models have been
developed to estimate contributions of these mechanisms to
the total backscatter. In this paper we investigate the
separation of scattering mechanisms for the classification
of sea ice. Airborne data from the JPL AIRSAR as well as
from the Canadian CV-580 is used. Based on the eigenvector
decomposition of the coherency matrix, we sort the three
resulting rank-one coherency matrices by the alpha angle
estimate of the eigenvectors (as opposed to sorting by
their corresponding eigenvalues). This sorting allows an
interpretation of the scattering mechanism involved, a
concept first introduced in . In a multi-stage approach,
we first focus on surface scattering, the main factor for
first year ice (FYI), and investigate how other
contributions (i.e. volume scattering and double bounce)
can then best be used to refine the classification result.
The multi-stage classifier results are compared to
single-stage classification results. Our main objective is
the classification of C-band polarimetric data, to develop
RADARSAT-2 applications. But as AIRSAR data are available
in C-, L-, and P- band, a multi-frequency approach will
also be discussed and will be used as reference result.
Scheuchl B., I. Hajnsek, and I. Cumming, 'Model based Sea
Ice Classification Using Polarimetric SAR Data,' Proc.
IGARSS 02, Toronto, 2002.