Noise Reduction Method Based on Principal Component Analysis With Beta Process for Micro-Doppler Radar Signatures

In radar remote-sensing area, the radar returns from a target are usually under relatively low signal-noise ratio (SNR) due to the large distance between radar and target, which will bring difficulties in target detection, tracking, and classification. In this paper, an efficient algorithm is proposed to denoise the returned micro-Doppler radar signals under low SNR conditions. The new algorithm develops a nonparametric extension to the principal component analysis (PCA) model with the Beta process (BP) prior. The BP is a fully Bayesian conjugate prior which allows analytic posterior calculation and straightforward interference. This proposed Beta process-based principal component analysis (BP-PCA) is utilized to model the returned micro-Doppler signals from airplane targets and ground moving targets with low-resolution radar, where the number of principal components in PCA can be selected adaptively with the BP prior-based Bayesian structure. Noise reduction is accomplished via reconstructing the echo within the subspace that composed of the selected principal components and discarding the residual noise subspace. We demonstrate the noise reduction performance of the proposed model with measured micro-Doppler data from some different kinds of targets. The experimental results are also compared with some other state-of-the-art approaches.