RCI-DUNet: Efficient Complex-Valued Deep Unrolling Network for Multi-View MIMO Radar Coincidence Imaging

In the application of radar coincidence imaging (RCI) sparse reconstruction with target multi-view imaging cell scattering intensity (ICSI) fluctuation, the traditional compressed sensing (CS) algorithms suffer from undesirable imaging quality and high computational complexity. This paper proposes a complexvalued deep unrolling network for RCI based on the fast iterative shrinkage-thresholding algorithm (FISTA), called RCI-DUNet. We first convert the complex-valued form of the radar signal to realvalued form using the complex-to-real arithmetic rule, and then propose a momentum correction module (MCM) with an acceleration function that achieves stable gradient propagation while maintaining phase coherence. We design a sparse transformation function based on convolutional neural network (CNN) and channel attention (CA) mechanism to dynamically adapt to target multiview ICSI fluctuation, effectively solving the problem of correlation mismatch between reference signals and echoes. We apply RCIDUNet to multipe input multiple output (MIMO) RCI of target multi-view ICSI fluctuation. It is shown that RCI-DUNet can provide high-quality imaging results even at low sampling rates with drastic ICSI fluctuation. Moreover, RCI-DUNet can greatly reduce imaging time compared to traditional CS algorithms, which can be used for fast RCI. The simulation results validate that RCI-DUNet achieves excellent reconstruction performance and high efficiency in MIMO RCI with target multi-view ICSI fluctuation.