Outlier-robust Bi-percentile Estimator Of The IG-CG Model In Sea Clutter And Sea-land-noise Ternary Scene Segmentation Method

Whether in the military field or civilian areas,target detection technology under the background of sea clutter plays a very important role.Aimed at the spatial non-heterogeneity and temporal non-stationary of sea clutter space,the accurate modeling of sea clutter is the key of target detection in sea clutter.Compound-Gaussian model with the inverse Gaussian texture(IG-CG)is recognized as one of the best models to describe heavy-tailed clutter by high resolution radars,and the estimation of the model parameters is a very hot research topic of radar signal processing,especially in the presence of abnormal samples in the real sea environment.When shore-based or airborne radar scans the sea surface,the echoes are often mixed with sea clutter,land clutter,island reef clutter,dominant noise and other areas,in which the returns of radar interference with different characteristics need to adopt different treatment methods for target detection.Therefore,according to the real sea data,the detecting scene is divided into land clutter,sea clutter and noise-dominated areas and ternary scene segmentation becomes the indispensable preprocessing step before target detection.In order to solve this problem,a new sea-land-noise ternary scene segmentation algorithm based on composite measure and morphological is proposed.The content of this thesis is arranged as following:Firstly,the physical mechanism of sea clutter and several common amplitude distribution models for sea clutter are mainly reviewed,and their application ranges as well as characteristics are analyzed.Then introduces the Compound-Gaussian model,and focuses on the properties of Compound-Gaussian model with the inverse Gaussian texture.Secondly,considering the fact that the presence of abnormal samples in the real sea environment has severe impact on traditional torque estimation and maximum likelihood estimation performance,an outlier-robust bi-percentile estimator of the IG-CG model is proposed.By simulated experiments,the bi-percentile estimator is studied on its performance with the number of samples,and is compared with moment estimator and maximum likelihood estimator when samples are with and without outliers,then the results show that the bi-percentile estimator with a good ability to resist abnormal samples performs better.Generally in the real radar environment,the returns of ships,reefs and sea spikes have much higher power than the average power of sea clutter and randomly appear at quite few samples.At last it is proved the robustness and accuracy of the bi-percentile estimation of parameters of IG-CG model in the real sea environment.Thirdly,a brief theory of image segmentation and application are introduced.Four classic image segmentation methods are reviewed as well as their advantages and disadvantages.And morphology in the image processing are introduced,including binary morphology and grayscale morphology as well as the application of morphology in the image processing,what builds a foundation for the research on scene segmentation methods in the next chapter.Finally,according to the distribution characteristics of the power of sea,land and noise,a new method of the sea-land-noise ternary scene segmentation with power and phase linearity mixed measure,which is based on the original sea and land segmentation method is proposed.And the segmentation result is verified by real sea clutter datasets.