High resolution ISAR imaging based on 2D-FMFSBL-net

In this article, a deep unfolded network, named 2D-FMFSBL-Net, is proposed for high resolution ISAR imaging. Firstly, the ISAR sparse observation model is established. Then the complex-valued 2D fast mean field sparse Bayesian learning(2D-FMFSBL) is derived to solve the model, and the iterative procedure of which is unrolled into a deep network, where all the adjustable parameters are learned through back-propagation. Since the optimal parameters of each layer can be learned separately, the proposed method exhibits more flexibility and achieves well-focused ISAR imaging with a shallow network. Experimental results on simulated and measured data demonstrate the effectiveness of the proposed method in incomplete data and low signal-to-noise ratio (SNR) scenarios.