Accurate approximation to the optimum parameter estimate for K-distributed clutter

The authors analyse the suboptimal performance of simple texture measures for estimating the order parameter of K-distributed radar clutter. A noncommittal neural net has been applied to the parameter estimation task and it shows that improved error estimates are obtained when multiple moments are used to characterise the texture. A new estimator is proposed which combines the mean normalised log intensity and the amplitude contrast moments of the imaged data to provide a more accurate measure of the texture information, which results in lower errors in the parameter estimates. The relative weighting in which the two moments are combined determines the error performance of the estimator. Using a constant weight value, an estimator has been derived which gives close to maximum likelihood performance on the estimates over a wide range of the parameter values which are of interest. Thus it is shown that the use of multiple moments in a texture measure produces a closer approximation to the optimum parameter estimate for a K-distributed process.