Study On Civil Radar Target Recognition Based On Micro-Doppler Signature

Different movement of different components makes the specific Doppler modulation in the whole echo signal,that is,micro-Doppler phenomenon.No demand for high range resolution,as well as the decreasing cost of millimeter-wave Radar and the rapid growth of the civilian radar market,makes the application of the micro-Doppler based Radar Automatic Target Recognition(RATR)technology for civilian automotive millimeterwave Radars,a new developing field and research hotspot,a good performance price ratio.Due to the difficulty in obtaining measured data,in this dissertation,we make a simulation study on civil application of micro-Doppler based target recognition in the field of Radar Signal Processing(RSP),focusing on key technologies such as feature extraction,statistical learning,noise reduction and signal reconstruction.The main work is summarized as follows:Two types of human-vehicle recognition processes are designed: the TimeFrequency Spectrum(TFS)based effective feature extraction method,Empirical Mode Decomposition(EMD)based feature extraction method,time-series feature extraction method,and the classification theory of Support Vector Machine(SVM)and the theory of Hidden Markov Model(HMM).Simulation results show that the target speed direction,speed size and dwell time have different effects on the classification of SVM and HMM.Under general conditions,the classification accuracy of different targets of SVM and HMM is above 95% and 99%,respectively.The noise reduction and signal reconstruction problem,which can improve precision and robustness under low Signal-to-Noise Ratio(SNR),are studied.Principal Component Analysis(PCA)is a classic dimensionality reduction denoising method based on the theory of Signal Space.Sparse Representation describes the signal as sparse linear combinations of a set of prototype signal-atoms,the overcomplete Dictionary,by some classic algorithms such as Basis Pursuit,Orthogonal Matching Pursuit,etc.The Dictionary can be a prespecified set of linear transforms or adaptive learnt from a set of training signals.Compressive Sensing(CS),which aims to recover signals from incomplete linear measurements,is based on the hypothesis that the signals have compressible representations as long as the Sensing Matrix satisfying the Restricted Isometry Property.Simulation results show that both PCA and PPCA-BIC can improve SVM classification accuracy of high dwell time samples and HMM classification accuracy on low signal-to-noise ratio condition.The PPCA-BIC has obvious advantages over PCA,and PCA may reduce the SVM classification accuracy of the low dwell time and high signal-to-noise ratio samples.SCS has good practical value.Reconstruction Results corroborate that when data missing rate is 0.6 or 0.8,and even lossless reconstruction can be achieved when the data missing rate is 0.8.An adaptive dictionary learning algorithm which implements the K-SVD,is hard for applying to processing high-dimensional signals.