Détection CFAR en Milieux Non-Gaussiens Inhomogènes

Abstract

In this Master’s thesis, we deal with the improvement of the performance of the CFAR (Constant False Alarm Rate) detection of a Swerling I type target embedded in an inhomogeneous Weibull or Log-normal clutter with unknown parameters. We assume that the reference window may contain a clutter edge, with an unknown position, partitioning the reference window into two sets of IID (independent and identically distributed) Weibull or Log-normal samples each but with different powers. In doing this, we analyze and compare two techniques of automatic localization of a clutter edge position [1-3] whose goal is the automatic censoring of the unwanted samples of either smaller or greater power. To this end, in the first technique [1], a logarithmic amplifier is necessarily introduced; i.e., the Weibull and Log-normal distributions are reduced, respectively, to a Gumbel and Normal distributions. In the second, however, technique found in [2, 3], the samples should first be sorted in an ascending order. Finally, to estimate the clutter level and decide the presence or the absence of the primary target, the selected homogeneous set yielded by each technique is incorporated in one of the suitable CFAR detectors found in the radar literature. More precisely, the CFAR detector should perform two tasks (algorithms); censoring and CFAR detection. Indeed, the virtue of the automatic censoring detector is the considerable improvement of detection performance over the corresponding detector, i.e., fixed point censoring detector. To assess the efficiency of the censoring and detection of each improved detector, we compare it in different scenarios, through Monte Carlo simulations, with its corresponding fixed point censoring detector. The obtained results show that both exhibit the same performance in a homogeneous clutter; however the former significantly outperforms the latter in the presence of a clutter edge within the reference window.