Study On Classification Of Airplane Targets Based On Micro-doppler Effect

When radar illuminates a moving target, the carrier frequency of the returned signal will be shifted. This phenomenon is known as Doppler effect. If the target or any structural component of the target has oscillatory or rotation motion, which is referred to as micro-motion, such micro-motion may induce the additional Doppler modulations around the target’s Doppler frequency, which is called as micro-Doppler effect. Micro-motion and corresponding micro-Doppler effect are highly related to the configuration structure and the motion characteristic of the target and are capable of providing us more detailed information about the target. Therefore, radar automatic target recognition(RATR) based on micro-Doppler effect is becoming a research field which has vast application prospect. This dissertation focuses on the three problems faced when utilizing the micro-Doppler effect to realize the RATR tasks via the low resolution radar, i.e., micro-Doppler feature extraction, target classification under the low Signal-to-Noise Ratio(SNR) cases and target classification under the short dwell time cases. The main content of this dissertation is summarized as follows: 1. A detailed analysis of the micro-Doppler modulation characteristics of some typical targets for low resolution radar is given. The main work includes:(1) For the airplane targets including turbojet aircraft, prop aircraft and helicopter, we first establish the mathematical model for the rotor blades in the aircraft. Based on the mathematical model, simulated data and measured data obtained via a low resolution radar are analyzed to demonstrate the distinctions of the micro-Doppler modulations between different kinds of aircrafts. The analysis results show the feasibility of classification of the three kinds of aircrafts using micro-Doppler signatures;(2) Since the forms of movements of the ground moving targets, i.e., single walking person, two people walking and a moving vehicle are complex and diversiform, we qualitatively analyze the micro-Doppler modulation characteristics of the three kinds of targets based on the measured data via low resolution radar. 2. Study on the classification scheme to categorize the airplanes into three kinds, i.e., turbojet aircraft, prop aircraft and helicopter based on the micro-Doppler characteristics of their low resolution radar returns. The main work includes:(1) A novel micro-Doppler feature extraction method based on Empirical Mode Decomposition(EMD) and CLEAN technique is proposed. Firstly, EMD is utilized to decompose the radar returns from the three kinds of aircrafts. The results based on measured data show that EMD can decompose the fuselage component and the micro-Doppler component effectively for the turbojet aircraft and helicopter. Since EMD failed to decompose the fuselage component and micro-Doppler component for a prop aircraft, CLEAN technique is further utilized to analyze the radar returns from the three kinds of targets. Based on the decomposition results of EMD and CLEAN, five dimensional features reflecting the energy ratio between fuselage component and micro-Doppler component or reflecting the different micro-Doppler modulations of different kinds of aircrafts are extracted for the classification. Experimental results based on the simulated and measured data show that the proposed EMD-CLEAN method can not only achieve a good classification performance but require the relatively low parameters of radar system;(2) Since the micro-Doppler components are easily contaminated by the noise in the low SNR cases, we proposed a SR-CLEAN method based on the signal sparse representation(SR) theory and CLEAN technique to denoise the test samples. Experimental results based on the simulated and measured data show that the proposed SR-CLEAN method can evaluate the classification performance of the EMD-CLEAN method under the low SNR cases. 3. Since the SR-CLEAN denoisng method requires a precise estimation of the noise variance which can hardly be obtained in some cases, we develop a new algorithm to denoise the returned micro-Doppler radar signals under low SNR cases. This new algorithm develops a nonparametric extension to the principal component analysis(PCA) problem with the Beta process prior. Then the Beta process principal component analysis(BP-PCA) model is applied to the low resolution radar returns for automatic choice of the number of principal components. Noise reduction is accomplished via reconstructing the echo within the subspace composed of the selected principal components without the knowledge of the noise variance. Experimental results based on simulated data show that the proposed BP-PCA method can automatically and precisely determine the size of the signal subspace. Experimental results based on measured data show that the proposed method can not only achieve the good noise reduction performances but also preserve the micro-Doppler signatures of the radar returns under the relatively low SNR conditions. 4. Study on the signal reconstruction methods for the signals with missing samples obtained via low resolution radar aiming at solving the problem that the classification performance deteriorates dramatically under the short dwell time cases. The main work includes:(1) For the low resolution radar system, of which the target echoes can be assumed to follow the complex Gaussian distribution. Based on this precondition, we firstly formulate the probabilistic model between the observed signal with missing samples and the unknown complete signal. Then the posterior distribution of the complete signal is obtained via the Bayes’ theorem. We obtain the maximum likelihood estimation of the model parameters using the Variational Bayesian Expectation Maximization(VBEM) algorithm, of which the posterior mean of the complete signal is our reconstruction.(2) Due to the fact that the low resolution radar echoes spread within some low dimensional subspace instead of the whole high dimensional subspace, we proposed an Automatic Relevance Determination prior based Factor Analysis(ARF-FA) model to reconstruct the signals with missing sample. FA is utilized to decompose the covariance matrix of the complex Gaussian distribution. Comparing with the complex Gaussian model, the freedom degree of the proposed ARD-FA model is much smaller which means in the reconstruction process the ARD-FA model needs less test samples. In addition, the ARD prior can automatically determine the number of factors in FA model. Experiments based on the measured data show that the proposed methods can achieve the good reconstruction performance compared with other state-of-art approaches...