Research On Estimation Of Parameters Of Sea Clutter Model With Log-normal Texture

Maritime radars inevitably receive the backscattered returns from the sea surface,which is referred to as sea clutter.Sea clutter is an important factor to affect the target detection performance of maritime surveillance radars.Accurate statistical modeling of sea clutter and precise parameter estimation are the preconditions for adaptive detection of targets in sea clutter.The compound-Gaussian model can accurately characterize the statistics of sea clutter,which represents sea clutter time sequences as the product of fast-varying complex Gaussian speckle component and slowly-varying positive texture component.Different types of texture distributions correspond to different compound-Gaussian models.The measured data show that the compound-Gaussian model with log-normal distributed texture(CG-LNT)can provide good goodness-of-the-fit to high-resolution sea clutter data.The estimation of the shape parameter and scale parameter of the model is of significance to maritime radar target detection.However,the existing moment-based methods have poor estimation performance and fail to meet the demand of adaptive target detection.Moreover,these estimators are quite sensitive to outliers,and their estimation performance degrades sharply in real clutter environment with outliers.To solve the problem,this thesis focuses on the effective estimation methods of the parameters of the sea clutter model with lognormal texture.The main content of this thesis is summarized as follows:In the second chapter,the basic theory of compound-Gaussian sea clutter model is introduced.Then,the widely used compound-Gaussian models with different texture distributions are introduced,and the statistical characteristics of different models are analyzed.Finally,the existing parameter estimation methods of different sea clutter models are reviewed and discussed.The third chapter focuses on the moment-based estimation methods.The existing momentbased estimation methods on clean sea clutter data have low precision for the sea clutter model with lognormal distributed textures.Here,the estimators based on adaptive fractionalorder moment and logarithmic cumulants are proposed.By improving the existing fractional moment estimation methods,the order of the moment used in parameter estimation is matched with the shape parameter,which effectively improves the estimation performance.Then,the theoretical formula of the logarithmic cumulant is derived through the Merlin transform.The parameter estimation is made by different logarithmic sample moments,and the orders of the moments are also matched the shape parameter.Finally,the performance of the proposed estimation method is evaluated by experiments using simulated and measured data.In the fourth chapter,in view of the fact that the parameter estimation method based on sample moments has poor performance in the case with outliers,a robust tri-percentile estimator is proposed.Firstly,the relationship between percentile ratio and shape parameter of CG-LNT clutter model is analyzed and established.The shape parameter can be estimated by two sample percentiles.Then,the scale parameter is estimated by the third percentile which depends on the estimated shape parameter.After that,the relevant properties of the tri-percentile estimator are analyzed,and the empirical formulas to optimize the three percentiles positions are proposed.Finally,the outlier-robustness of the proposed tripercentile estimator is verified by experiments using simulated and measured data.Finally,the works in the thesis are summarized and the future works in this field is discussed.