Investigation into the use of nonlinear predictor networks to improve the performance of maritime surveillance radar target detectors

Previous research has claimed that sea clutter is a chaotic process with a nonlinear predictor function. Indeed, results have been reported which demonstrate that sea clutter is nonlinearly predictable, and that this predictability can be exploited, using nonlinear predictor networks, to improve the performance of maritime surveillance radars. The aim of the paper is to investigate if nonlinear predictor networks can be used to improve the performance of maritime surveillance radars, using sea clutter data sets provided by the Defence Evaluation and Research Agency (DERA). By presenting prediction results for radial basis function network predictors, Volterra series filter predictors, and linear predictors, it is shown that the clutter predictor functions are well approximated by a linear function, and that nonlinear predictor networks provide little or no improvement in performance. A novel and effective training methodology is used for the radial basis function network predictors