Research Of Surveillant Radar Target Classification Method Based On Micro-Doppler Signature

Surveillant radar have the ability of searching targets in a wide and it widely applicated.But constrainted by its task and the system limitation,surveillant radar tent to have low resolution,low pulse rate frequency and short radar beam dwelling time.Therefore,it is difficult to obtain more information of targets in detail.The micro-Doppler signal contain in the target radar echo can reflect certain target motion information.By analyzing the features of these signals,surveillant radar can simply classify the target and evaluate the threat while accomplishing the target detection and search.At present,most of the research on micro-Doppler signal is carried out by time-frequency analysis method.In this thesis,to satisfy the special condition of surveillance radar a micro-Doppler signal classification method based on statistical features is proposedThe innovative part of the paper includes the method of filtering the target bulk echo from the micro-Doppler signal in a combination method of L-statistics algorithm and CLEAN algorithm.The method can remove the echoes of the moving object with unconstant Doppler frequency and keep the micro-Doppler signal which can only produce slight Doppler shift well.The validity of the method is validated by simulated micro-Doppler signal.In this paper,we propose a method to improve the standard generalized edit distance(NGLD)in string matching so that it can reflect the similarity between two signals,and search the micro-Doppler signal period with NGLD as the cost function.The characteristics of NGLD which are used to reflect the similarity between signals and the parameter selection strategy of NGLD,are verified by Monte Carlo simulation method.The validity of the micro-Doppler period searching method based on NGLD is verified by the simulation data and the measured data both.And the results of this method are compared with those methods using the similarity degree of envelope and time-frequency graph as the cost function respectively.The results show that,the method based on NGLD is superior to the other two methods.On the basis of obtaining the signal micro-Doppler period,six features extracted from radar echo signals in time and frequency domain are used to train and test an Adaboost classifier.The experimental results show that the error probability of the proposed method is only 3.75% for the measured propeller model and wheel model.