Estimation et détection des signaux radar pour la surveillance maritime.

Abstract

Appropriate techniques for radar signal processing operating in a maritime environment may be substantially enhanced through a deep understanding of the direct target and sea surface backscatters. These latter are usually referenced as sea clutter, inevitably encountered during the detection process. Except for a few applications, sea clutter represents the unwanted part of the retransmitted signal which may be harmful to target detection. It is shown through the analysis of real maritime data that the intensity properties of HRR (High Resolution Radar) backscatters can be represented by the non-Gaussian model of the Pareto distribution, which has the characteristic of having a simple and powerful mathematical model compared to a wide range of techniques from other long-tailed distributions. In our thesis, we are interested in studying data modeled by the thermal noise free Pareto distribution. Our work is divided into two main parts. The first deals with the study of parameter estimation using the VSE (V-Statistic Estimator) and the USE (U-Statistic Estimator), based on the fractional negative and positive order moments, and the COSE (Consecutive Order Statistic Estimator), based on a consecutive order statistic technique of the Pareto distribution modeling the maritime clutter, in which the target is embedded. The second considers HRR target detection through adaptive threshold and CFAR (Constant False Alarm Rate) algorithms, by exploiting a combination of samples collected from both sides of a target. In this context, the twofold ACD-CFCAR (Automatic Censoring-Constant False Censoring and Alarm Rates) processor; that is, automatic censoring and detection, the EVI-ASDCFAR (Enhanced Variability Index- Automatic Selection Detector-Constant False Alarm Rate) and the PI-CFAR (Pietra Index-CFAR) intelligent processor which achieves automatic switching between several detectors, i.e., GM- (Geometric Mean-), GO (Greatest Of-) or TM (Trimmed Mean- CFAR), are proposed. In doing this, we treat, through Monte Carlo simulations, the problem of the best possible detection performance of Swerling I type targets embedded in a Pareto clutter. The results are such as the ACD-CFAR detector exhibits good censoring and detection performances in multiple target situations because, each time, it censures the exact number of interfering targets comparatively to the conventional TM-CFAR whose performance relies upon the a priori known environment. Then, we introduce automatic target detection in homogeneous and heterogeneous Pareto clutter. To this end, we develop the EVI-ASD-CFAR processor. In effect, in an attempt to use the VI (Variability Index) as a clutter discriminator, we examine this processor in the presence of multiple targets. A dynamic selection of the appropriate detector among the GM-CFAR family or the TM-CFAR detector is carried out through the Pareto-exponential duality relation, and a switching logic based on the VI and MR