Research On Prediction Method Of Key Characteristic Parameters Of Sea Clutter

The estimation of sea clutter characteristic parameters and the determination of amplitude distribution type of sea clutter data are two crucial issues in the field of sea clutter characteristic perception and suppression.The reflectance of sea clutter is a key characteristic parameter that can directly reflect the power of the sea clutter.Its high-precision prediction is of great importance.However,the existing prediction methods are mainly based upon several classic empirical models.The empirical models mainly focus on the wide coverage on frequency bands and universality and often have low prediction precision.The low precision fail to meet the personalized needs for some radar systems with subtle signal processing.In addition,sea clutter can be modeled by the compound-Gaussian models of different types of texture distribution.In applications,it is an important problem that must be first solved how to determine the best type of texture distribution from oceanic environment parameters(sea state),radar parameters,and the viewing geometry of the radar.At present,commonly-used methods are all qualitative empirical criteria.Faced with complex and varying large scene sea environments,these methods are not enough ability to determine the best type of texture distributions.Aiming at the two important problems in the perception of sea clutter characteristics,the thesis investigates the methods to predict the key characteristic parameters of sea clutter and to determine the type of the texture distributions from the sea environment parameters,radar parameters,and the viewing geometry of the radar by means of neural networks and classification learning networks.At last,the thesis proposes the prediction method of the reflectance of sea clutter using neural networks and the determination method of the type of texture distributions using the support vector machine(SVM).The main content of the thesis is summarized as follows:In the first chapter,we introduce the research background of the work in thesis and the organization of the content of the thesis.In the second chapter,the relevant knowledge on sea clutter characteristics and neural networks are introduced.Firstly,the scattering characteristics of sea clutter are reviewed.And then,three types of texture distributions in the compound Gaussian model of sea clutter are briefly introduced.At last,two kinds of commonly-used neural networks for nonlinear function approximation are reviewed.In the third chapter,the empirical formulae to predict the reflectance of sea clutter are reviewed.In view of low precision of these empirical formulae,the prediction method of the reflectance of sea clutter using neural network is proposed,which combines the effective factors of the empirical formulae with the learning ability of neural network.Moreover,the performance of several different combinations in the prediction neural network are contrasted by using measured sea clutter database to find the best combination.At last,a demonstrative software system for prediction of the reflectance of sea clutter is installed to visually illustrate the performance of different methods.The fourth chapter focuses on the determination of the best type of texture distributions of sea clutter by the classification neural network.Firstly,several commonly-used classification algorithms are introduced.Secondly,existing best type selection methods are briefly reviewed,including some qualitative criteria and the best type selection based on the K-S distance and the empirical CDF of data.At last,a method to determine the best type from sea environment parameters,radar parameters,and the viewing geometry of the radar is proposed,which is based upon the support vector machine(SVM).In the fifth chapter,our work in the thesis is concluded and the future work on this field is discussed.