Multi-feature Based Detection And Classification Of Small Targets On Sea Surface

With the continuous improvement of radar detection technology and the expansion of its application field,the accuracy of radar target detection algorithm is also improved.Due to the irregular and disordered characteristics of sea clutter,the ocean signal is also nonlinear,non-stationary and non-Gaussian,which leads to the failure of conventional functional models to fit the actual sea clutter signal.The method of building an appropriate functional model is not ideal in dealing with high sea conditions,so many scholars begin to extract features by means of signal processing to describe the features of sea clutter in one or more angles,and then put forward excellent feature-based sea target detection algorithm.At the same time,with the development of machine learning technology,more and more scholars integrate machine learning algorithm to conduct intelligent processing and research on sea clutter and targets.The purpose of this thesis is to study the characteristics of the measured radar data at various angles and to propose an effective detection algorithm and classification method for sea surface targets.Firstly,from the perspective of time-frequency analysis,the measured data will be decomposed and features will be analyzed by Empirical Mode Decomposition(EMD),and the detection of small targets on the sea surface are completed according to the detection theory.At the same time,the concept of intelligent processing is introduced to extract effective features,and an effective classification algorithm is proposed to complete the classification of sea clutter and targets.The research contents of this thesis are summarized as follows:Firstly,the definition of EMD decomposition and the properties and characteristics of EMD are introduced,and the decomposition process is completed according to the measured radar echo data.At the same time,the advantages and disadvantages of EMD decomposition in dealing with nonlinear and non-stationary complex sea clutter signals are analyzed by using the results of measured data.Then,Intrinsic Mode Function(IMF)components are decomposed by EMD to extract energy proportion features,which are used as detection statistics to complete detection of small targets on the sea surface.Correlation coefficients of IMF components and received echo data were established respectively,and the ratio of mean to standard deviation is used as the criterion for screening IMF components,so as to automatically screen out low-order IMF components with large energy and stable fluctuation.The average energy ratio of IMF component in the original signal is extracted as the feature to detect the anomaly of target on the sea surface,and compared with the correlation detection algorithm.Then,we will extract features from multiple angles,fuse multiple features,and display and analysis feature.By analyzing the Krogager polarization decomposition,the relative power of sphere,biplane and helical scattering are extracted as the research features.Then,the characteristics of relative mean amplitude and non-extensive entropy are extracted from the perspectives of time domain and frequency domain.We will contrast and analyze the difference between sea clutter and target echo.Finally,based on the extracted features,a multi-feature joint classification method under sample imbalance is proposed.In order to solve the problem of sample imbalance and feature aliasing,a kind of classifier combining K-means and support vector machine is designed.The classifier mainly conducts K-means dynamic clustering of sea clutter samples,divides the sea clutter samples originally belonging to one class into multiple classes,and then classifies the multi-class sea clutter samples and target samples by Support Vector Machines(SVM).This method can effectively classify sea clutter and targets.