Time-Varying Autoregressive Modeling of High Range Resolution Radar Signatures for Classi cation of Noncooperative Targets

Time-varying autoregressive (TVAR) modeling approach for the representation of complex non-stationary process is presented, and applied to the classiication of High Range Resolution (HRR) radar signatures. HRR radar signatures are multi-channel nonstationary complex processes, and features extracted under the stationary assumption often result in unsatisfactory performance. In a TVAR modeling approach, the TVAR coeecients are expanded as a linear combination of deter-ministic time functions. In this paper, the TVAR coeecients are expanded by a low-order discrete Fourier transform (DFT), since the TVAR parameters are complex and need be expanded by a set of complex functions. The parameter estimation and order selection in TVAR models are also considered. A least squares estimator of the TVAR model parameters is presented, and the maximum likelihood approach for determining the model order is also presented. The eecacy of the TVAR modeling approach in modeling nonstationary processes is demonstrated by estimating the time-varying spectrum of a synthetic process by diierent approaches. For the classiication of HRR radar signatures, the estimated TVAR model parameters are used as features. A neural network is used to classify TVAR features extracted from HRR radar signatures. Two sets of non-cooperating target identii-cation (NCTI) data, each set contains 2500 training samples and 2500 test samples in ve classes, are used in the experiment. The neural network is trained with training samples in ve classes, and is used for classifying test samples. In the classiication experiment, about 93% of samples in the better-aligned NCTI data set are correctly classiied.