Characterization of spatial statistics of distributed targets in SAR data

Abstract

A method for the analysis of spatial statistics in multifrequency polarimetric Synthetic Aperture Radar (SAR) data is presented. The objective is to extract the intrinsic variability of the target by removing the variability from other sources. Three sources which contribute to the spatial variability in the returned power from a distributed target are modelled, they are (1) image speckle, (2) system noise, and (3) the intrinsic spatial variability of the target or texture. Speckle and system noise are modelled based on an understanding of the physics of the SAR imaging and processing systems. Texture is modelled as a random variable which modulates the mean returned power from a distributed target. An image model which accounts for all three sources of variability is presented. The presence of texture is shown to increase the image variance-to-mean square ratio and to introduce deviations of the image autocovariance function from the expected SAR system response. Two textural parameters, the standard deviation of texture and its autocovariance coefficient, are examined. This statistical approach is illustrated using sea-ice SAR imagery acquired by the Jet Propulsion Laboratory three-frequency polarimetric airborne SAR. Textural modulation of the signal has been detected in the imagery. Results show that for different sea-ice types the spatial statistics seem to vary more across frequency than across polarization and the observed differences increase in magnitude with decreasing frequency. The results also suggest the potential of this approach for discrimination of various sea-ice types and open water in single frequency, single polarization SAR data. Correlation of the spatial statistics to field measurements will be important for the verification of these observations.