Micro-motion Feature Extraction Based on Bayesian Inference

The micro-Doppler (m-D) effect caused by periodic Doppler modulation of the micro motions such as rotation, vibration, coning motion and precessional motion is widely used in target recognition by estimating the m-D parameters. In this paper, the echo signal of micro-motion targets is projected on the m-D parameter domain to obtain the sparse representation and the m-D parameter estimation can be regarded as the problem of sparse signal recovery. To estimate all three m-D parameters efficiently, a novel parametric Bayesian compressive sensing (BCS) reconstruction algorithm is proposed, in which the Doppler repetition period and Doppler amplitude are discretized to structure an over-complete dictionary and the initial phase estimation is transformed into a dictionary mismatch problem, which is solved by the sparse Bayesian inference. The effectiveness of the proposed method is validated by simulations.