Gouri Amel2021-01-172021-01-172020http://dspace.univ-msila.dz:8080//xmlui/handle/123456789/23320Compound Gaussian (CG) models with different texture components have been utilized to describe sea clutter statistics at low grazing angles. In the first work, the well-known K, Pareto type II and compound Gaussian inverse Gaussian (CGIG) models are combined firstly to produce mixture distributions with thermal noise. Using Intelligent Pixel Processing (IPIX) real data, the fitting of the proposed models are investigated where a simplex algorithm based on the Nelder-Mead algorithm is used to achieve the unknown parameters from the data. Regardless of heavy computation time, fitting results to empirical data showed the efficiency of the proposed mixtures models against the standard distributions in several cases. The second work considered in this thesis is the development of new estimation procedures labeled higher order moments estimator (HOME), constrained non-integer order moments estimator (CNIOME) and constrained maximum likelihood estimation (CMLE) for parameter estimation of CGIC clutter plus noise. The HOME method is given in a closed-form for both known and unknown clutter-to-noise ratio (CNR). As the last two estimators are given in integral forms, the Gauss quadrature method based on Legendre and Laguerre polynomials is used after splitting the underlying integrals into two parts. Using simulated and real data, estimation performances of the proposed methods are assessed in terms of known and unknown clutter to noise ratio. In the case of known CNR, it is shown that the HOME and the CMLE methods can produce similar estimates of the shape parameter at high sample sizes. For unknown CNR, both CNIOME and CMLE methods present approximate results. Moreover, the HOME exhibits poor estimation results for all values of the shape parameter. When the number of integrated pulses and CNR values are high, the CMLE method outperforms always the CNIOME and the HOME methods. In radar target detection, parametric and non-parametric Constant False Alarm Rate (CFAR) detectors for Weibull distributed clutter have been suggested achieving the full CFAR property. The performances of these CFAR algorithms are measured by CFAR loss in homogeneous and heterogeneous environments. The improvement of detection performances in terms of low time-consuming which is very important in real time applications is also considered. To this effect, the zlog(z) based estimator for CFAR detection in homogeneous Weibull clutter is used. This estimation method is obtained in terms of the digamma function where the estimates of the shape parameter are determined by the interpolation tool. The non-integer order moments estimator (NIOME) is also given and coincides the zlog(z) estimation results for low values of the moment’s fractional order. Based on the Neyman-Pearson type test, CFAR detection comparisons, based on the proposed estimator and the existing logt-CFAR and maximum likelihood CFAR (ML-CFAR) detectors, are conducted.Thesis