High Frequency Surface Wave Radar Complex Clutter Recognition Based On Machine Learning Technology

High frequency ground wave radar(HFSWR)can detect distance target beyond the earth curvature based on the interaction between 3M-30 M high frequency vertical polarization wave and the sea surface,which is of significant importance to our coastal and national security,to navigation safety,and to environmental monitoring.However,HFSWR is interrupted by various kinds of clutter,such as sea clutter,ionospheric clutter,environment noise,meteor,and radio interference.All of these lead to a great impact on target detection for HFSWR.The current main clutter suppression methods including space-time processing,time-frequency processing,image processing,clutter modeling analysis,each method is linked to a specific kind of clutter on a certain condition.If we can adaptively discriminate different clutter in range-doppler(RD)spectrum of HFSWR,not only can clutter be sent to their exclusive suppression module,but also can the target detection strategy be adaptively shifted,which give the radar processing a more intelligent mode.The final purpose of the paper is to give a better way to discriminate various clutter in HFSWR.As the border of various clutter in the actual RD spectrum data is unknown,we first simulated the RD spectrum based on the sea clutter generation mechanism,radar equation,ground wave attenuation principle and the statistical properties of ionospheric clutter.This paper will introduce the theory to make clutter simulations as well as their labels under complex environment.The recognition and suppression of clutter is always the focus of the research.We first establish a clutter feature library based on recent knowledge of clutter,then we successively extract the power feature,dimensionless shunting feature,wavelet feature,gabor feature,statistical feature,theoretical position feature,signal-clutter ratio feature and a united feature.We also analysis and select the adequate feature by means of feature statistical,information theory and classification performance.The best feature combination is selected to a clutter classification module based on support vector machine,which achieved a satisfactory performance.A clutter classification method based on convolutional neural network(CNN)is presented in this paper.The preprocessing and the input sub-image is introduced primary.After the training process in CNN,the performance of the net is compared with the selected feature classification method.The method based on CNN also gain a well performance,which provides a new way for the classification of clutter.Finally,the proposed method is applied on real data.We put forward an automatical classification method for a quantilization evaluation.We also provide a GUI(Graphical User Interface)for further completion.The proposed method gains feasibility of practical application as well.