A Research On Parameter Estimation Of Sea Clutter Magnitude Distribution

The accurate and physical motivated modelling of radar returns from sea surface provides a firm foundation on the improvement of performance of maritime radars,whose parameters need to be estimated precisely to achieve the adaptive detection of radar targets.Owing to its simplicity,the compound Gaussian model,with three different textures,has been used widely to describe the nongaussianity of sea clutter.A number of estimators have been provided,without giving a satisfied solution of the following problems.Firstly,although the classic maximum likelihood estimator achieves a nearly optimum performance without prior knowledge,it can be hardly used in real radar system due to its high calculation complexity.Besides,traditional estimators are sensitive to the outliers,a kind of high power data widespread in the radar returns with a remote possibility.What is more,faced with“small sample problem”,the lack of independent identically distributed data samples,which is caused by the diversity of offshore area under high resolution radar,traditional methods barely give a satisfied result for later target detection.This thesis focuses on the parameter estimation problem of compound Gaussian model,whose major contributions can be summarized as follows:A brief introduction of the electromagnetic scattering of sea surface has been given in Chapter2,in which the classic models to describe them are listed.Following the detailed derivation of compound Gaussian model,we have listed a series of methods published for obtaining the parameters of those model.Here we provide an estimator,called the iterative maximum likelihood method,to improve the efficiency of the traditional maximum likelihood one.In Chapter3,we analyze the influence of the outliers which exist in the real scene,and introduce some solutions to this kind of problem.Since the common estimators can't offer a good way to avoid the impact of outliers,we have proposed a kind of parameter estimation based on the truncated distribution of the probability density function of data.Regarding as a generalization of the method based on bipercentile,the estimator can attain a robust performance while data includes a tiny number of outliers.In Chapter4,we propose a series of framework to estimate the parameters from data based on the Bayesian theory.Firstly,we generalize the zlogz method by using the fractional moment of zlogz called z Flogz method.Then,we find that the optimum order of z Flogz can be related to the shape parameter itself.By giving the empirical formula to fit the relationship between the shape parameter and the optimum order,we provide a method using the prior knowledge of shape parameter from multiscan data.Then after giving a brief introduction of the so-called “small sample problem”,we provide a Bayesian method using the statistical information of parameter,trying to give a reliable solution of this kind of problem.