Research On Estimation And Neural Network Prediction Method For Parameter Of Sea Clutter Model

Echoes received by a maritime radar are usually either backscattered signals from the sea surface or it plus target returns.The former is referred to as sea clutter.Maritime radar target detection is to judge whether radar echoes at a spatial resolution cell is sea clutter or sea clutter plus target returns.Sea clutter can be modeled by the compound Gaussian model,and the optimal and near optimal detectors have been developed for different types of compound Gaussian models.Therefore,estimation of the model parameters of sea clutter becomes an important problem in maritime radar target detection.At present,various moment-based estimators and the maximum likelihood estimator are rather sensitive to outliers of large amplitude in data.However,practical radar data inevitably contains some outliers that are probably from returns of targets or reefs.In view of this problem,this thesis carries out the research work at two aspects.On the one hand,the robust estimator based on percentiles for K-distributed clutter model is investigated,which can be used for parameter estimation and data annotation of sea clutter in parameter prediction from physical conditions to inquire the data.On the other hand,a prediction method of K-distributed parameters of sea clutter from physical parameters such as sea states and radar operating parameters is developed based on neural network learning.The main content of this thesis is summarized as follows:In the second chapter,the basic concepts and structures of the compound Gaussian model of sea clutter and neural network learning are introduced.Firstly,the problem of parameter estimation of compound Gaussian clutter is outlined.Parameter estimation problem of sea clutter is reviewed,and several existing methods of parameter estimation under different types of compound Gaussian models are introduced.At last,the basic principle and structure of the neural network for nonlinear function approximation are reviewed.The third chapter is aimed at parameter estimation of K-distributed sea clutter model,a new parameter estimator robust to outliers is proposed.By deriving the formula of the cumulative distribution function,the previous bi-percentile estimator is improved into a tri-percentile estimator,which markedly improves the estimation accuracy of scale parameter.Moreover,the performance of the tri-percentile estimator is evaluated by simulation data and measured data.The fourth chapter focuses on the neural network prediction of the parameters of K-distributed sea clutter from sea state,the radar operating parameters,and the viewing geometry of the radar using the recognized X-band CSIR radar database and a P-band radar database.Firstly,the data structure is analyzed and the radar or meteorological parameters which may affect the prediction of the parameters are extracted and normalized as the input data of the neural network.Then,the expected output of the parameter is obtained by the outlier-robust estimators from data,and the labeling of neural network's learning data is finished.By the learning to the annotated data,the prediction neural network is determined.Compared with existing empirical formulas,the neural network predictor attains much higher accuracy.