A Bayesian Framework for Detector Development in Pareto Distributed Clutter

Analysis of high-resolution X-band maritime surveillance radar clutter has demonstrated the validity of the Pareto Type II fit to such data. Based upon this, and the fact that in some cases the Pareto Type II model can be approximated by a Pareto Type I, it has been possible to design sliding window detectors with the constant false alarm rate (CFAR) property, concerning at least one of the Pareto model parameters. In the Pareto Type I case, it has been shown that CFAR can be achieved concerning both distributional parameters. For the case of Pareto Type II distributed clutter, only CFAR has been achieved concerning one of the clutter model parameters. To achieve full CFAR, in the Pareto Type II case, a novel Bayesian methodology is introduced. This approach is general and can be extended to other distributional settings. In the first instance, the Bayesian approach is outlined for single parameter clutter models, and then specialised to the exponentially distributed clutter setting. The method is then extended to two parameter clutter models, with particular focus on the Pareto Type II case. Jeffreys priors are applied in all cases.