Systèmes de détection de cibles réparties en milieu non gaussien.

Abstract

In this PhD thesis, we address the problem of detection of spatially distributed targets embedded in a non-Gaussian clutter. Since, in high resolution detection systems (HRR), the target is modeled as a set of dominant reflectors according to the ""MDS"" (Multiple Dominant Scattering) concept, we propose to design detection architectures that operate in non-Gaussian environments modeled by distributions such as: the K distribution, the compound Gaussian model with Inverse Gamma texture and the Pareto distribution .We first introduce a detection approach that detects MDS type targets embedded in a partially correlated distributed K environment whose parameters are unknown. This detector is referred to as M-pulse CA-LTCFAR (Multiple-pulse Cell Averaging based on Lookup Tables) . It is based on the integration of M-pulses, the CA detector and the use of Lookup tables (LT: Lookup Tables) and the integration of multiple pulses. This detector operates according to two essential phases: empirical computing of thresholding factors that maintain a Constant Pfa (Probability of False Alarm), and a phase of ""pulse-to-pulse"" parameters estimation. We also propose an expression of the total energy of the target after pulse integration, and construct from this expression, the statistical hypothesis test of the M-pulse CA-LT-CFAR detector. In the same context, we propose two mean level based on Lookup Table detectors , namely: the GO-LT-CFAR (Greatest Of Based on Lookup Tables) and SO-LT-CFAR (Smallest Based Lookup Tables). These two approaches are designed to detect MDS type targets embedded in compound Gaussian clutter with Inverse Gamma texture with unknown parameters. From the expression of the total energy of the target, we construct the statistical hypothesis tests of the GO-LTCFAR and the SO-LT-CFAR detectors. In addition, we introduce a detection approach that is based on the Geometric Mean (GM), allowing the detection of MDS targets embedded in a clutter modeled by the Pareto distribution. Also, based on the properties of the Pareto and Exponential distributions, we present the working principle of the GM detector for distributed targets, and derive an expression of the total energy of the target. Finally, We construct the statistical hypothesis test of the GM detector for distributed targets and propose a mathematical expression of the Pfa of the GM detector.