Research On Robust Parameter Estimation Methods Of Sea Clutter

Maritime radars inevitably encounter backscattered returns from the sea surface,referred to as sea clutter.In various maritime radar detection work,accurate perceptron theory over sea clutter characteristics based on matched statistic models is a pivotal basic to realize effective target detection and tracking.In maritime radar operating scene at high resolution and low grazing angle,it is an effective way to describe statistic characteristics of sea clutter echoes by applying various compound-Gaussian models with different distributed textures.Particularly,the three commonly-used biparametric clutter models are the K-distributed model,the generalized Pareto distribution model(GPDM)and the compound-Gaussian model with inverse Gaussian texture(IG-CGM).In practical application,the characteristic parameters of sea clutter models are supposed to be estimated by particular estimation methods,but there exist many problems in model parameter estimation due to complicated and various sea clutter environments,which effect the performances of parameter estimation and target detection severely.Particularly,several sea clutter characteristic estimation methods are proposed in this paper,which adapt to two types of radar work mode(localscene dwelling mode,large-scene fast scanning mode),three types of sea clutter statistical models(K-distributed model,GPDM,IG-CGM)and three types of sea clutter estimation environments(without outliers,with outliers,with and without outliers),respectively.The research results can be applied in many maritime radar detection scenes,such as antiinvasion,coastal surveillance and anti-submarine radars,in order to improve target detection performance of maritime radars.The key research and contributions in this dissertation are summarized as follows:1.In terms of the fact that the traditional high-order moment-based estimator of the IG-CGM in pure sea clutter environment is of low accuracy and unstable performance,the low-order moment-based parameter estimator and the iterative maximum likelihood estimator are given.On one hand,the low-order moment-based estimator is constructed by replacing the fourth-order sample moment in the traditional estimator by the first-order sample moment with higher accuracy and better stability.On the other hand,a computation-achievable and fast-converge iterative maximum likelihood estimator is derived by the partial derivatives of the logarithmic likelihood function related to two parameters.The results based on simulated and measured data demonstrate that both estimators can realize higher precision and more stable performance in parameter estimation in sea clutter environments without outliers,compared with the traditional method of moment.2.For the performance loss in characteristic parameter estimation caused by outliers existing in practical sea clutter estimation environment,explicit outlier-robust bipercentile(Bi P)estimators are proposed to estimate the scale and shape parameters of the GPDM.Based on the fact that the percentiles of sample vectors are robust to outliers of very small or large value,several explicit Bi P estimators are given based on the root formulae of the low-degree algebraic polynomials in some special setups of percentiles,and a fast-iterative formula is given to estimate the two parameters beyond these special setups of percentiles.Moreover,simulated and measured clutter data are used to verify the ability of the Bi P estimators in cases with and without outliers,and they are much better than the existing estimators with outliers in samples.3.In view of the fact that the outlier-robust explicit Bi P estimators suitable for the sea clutter environment with outliers cannot provide high-precision parameter estimation in the pure sea clutter environment,two modified parameter estimators are proposed from different aspects,including the combined bipercentile(CBi P)parameter estimator and the Bi P parameter estimator of bias reduction(Br-Bi P)based on the GPDM.First,considering that the explicit Bi P estimators are constructed by only a pair of percentiles in sample series,the CBi P estimators are derived by combining several pairs of percentiles with same parameter setup,and they are of higher estimation accuracy.Besides,the percentiles are consistent with respect to sample size,which causes performance loss in explicit Bi P estimators since only limited samples are available practically.In this paper,the bias properties are proved,and the Br-Bi P estimators are derived based on the bias indicative functions.Both the CBi P and Br-Bi P estimators are verified by simulated data and measured sea clutter data,and they can degrade the performance loss of the Bi P estimators in sea clutter environments without outliers effectively,and remain the robustness of the Bi P estimators in the case with outliers as well.4.For the robust tri-percentile estimators used for the K-distributed sea clutter in the case with outliers,the theoretical proofs of related properties and discussion and analysis of experiment performance are completed.On one hand,the effectiveness of the tri-percentile estimators and the asymptotical property of estimated error are proved in terms of relevant conclusions on sample percentile.On the other hand,simulated experiments are built to compare and analyze the performance of the tri-percentile estimators and existing estimators,and the simulated sea clutter and raw clutter data with some outliers are used to demonstrate the robustness of the tri-percentile estimators to outliers.The discussions show that the tripercentile estimators are the best choice when estimation precision,outlier-robustness,and computation time are all considered in the K-distributed sea clutter with outliers.5.Due to the spatial-temporally varying characteristics of sea clutter and spatially small sample problems of parameter estimation in large-scene scan-mode maritime radar detection,the spatial-temporally varying GPDM(STV-GPDM)and its fully adaptive coherent detection scheme of maritime radar are constructed,and the multiscan recursive Bayesian parameter estimation theory is proposed for the scheme.Firstly,the STV-GPDM model is constructed for large-scene clutter characteristic description,via replacing the fixed shape and scale parameters in traditional GPDM by dynamic parameter distributions related to spatial position and scan time.Secondly,a cognitive detection scheme of small targets in large-scene sea clutter is presented which consists of a characteristic perceptron of largescene sea clutter,a fully adaptive detector,and a percolator of abnormal samples.Finally,due to the lack of effective samples for parameter estimation,a multiscan recursive outlierrobust Bayesian estimation method is proposed for the sea clutter characteristic perceptron in the underlying scheme,based upon the idea of “remember only information but not record data”.Results via simulated and measured data show that the STV-GPDM model can describe clutter characteristics in large scene more subtly than traditional GPDM,and the fully adaptive coherent detection scheme based on the multiscan recursive outlier-robust Bayesian estimation method can provide better performance in parameter perceptron and target detection.