A Multibaseline Clutter Suppression Approach Assisted by Online Classification for Spaceborne Distributed MIMO Radar Systems

Spaceborne distributed multitransmit multireceive (also multiple-input multiple-output) (SD-MIMO) radar has great potential to enhance the capability of ground moving target indication (GMTI), particularly with the increasing of space-time-frequency degrees of freedom (DOFs). GMTI performance is, however, severely limited by complex geographical clutter. In order to address this issue, we proposed a clutter suppression method assisted by online classification using the range-Doppler-band 3-D dataset of SD-MIMO radar. The core points and innovations of our method are summarized as follows. First, we extend Markov random field (MRF) classification to 3-D space by integrating a 3-D adaptive weighted penalty (3-D-AWP) function and a probability distribution model of multichannel clutter. The 3-D-AWP function is constructed using a novel spatial distance metric designed to characterize the interaction potential of image pixels within their 3-D neighborhood. This ensures that the MRF smooth factor can be adaptively adjusted in both homogeneous and inhomogeneous clutter regions through different weighting values. The probability distribution model of clutter is established by the K-Wishart distribution, which incorporates amplitude and interferometric phase between multiple spatial channels. Then, under the Bayesian framework, we achieve the classification of multiband synthetic aperture radar (SAR) images by solving the maximum posterior probability problem. Subsequently, the homogeneous and inhomogeneous regions are identified based on classification results. Adaptive clutter suppression processing is performed independently for each subregion. Finally, the SAR-GMTI experiments are performed using synthetic data and measured data. The results demonstrated an improvement in clutter suppression capability by 2 dB and an enhancement in the target output signal to clutter plus noise ratio (oSCNR) by 1.5 dB compared to traditional methods.