Maximum likelihood estimation of point scatterers in synthetic aperture radar data

The motivation for this research is the recognition of objects from the relative locations of radar point scatterers on the object. Point scatterers are modeled as damped exponential signals. Parametric estimation of the exponential parameters determines the point scatterer locations. Maximum likelihood estimation of exponential parameters in white Gaussian noise provides good results in many exponential estimation problems. In fact, the exponential parameters are estimated with touch better than Fourier resolution. In the case of SAR, however, the white noise is estimated to be quite large even in relatively noiseless images. This is due to energy that is not well modeled by damped exponentials. This unmodeled energy distorts the estimated point scatterer locations. This paper proposes that the unmodeled energy be modeled as unknown colored noise and that maximum likelihood estimation of scatterer locations in unknown colored noise will provide more accurate scatterer locations. New methods for estimating exponential parameters in unknown colored noise are applied to SAR data. These methods efficiently model the points scatterers in SAR data. The new methods achieve a higher estimation accuracy than maximum likelihood estimation of exponentials in white noise.