Enhanced Through-the-wall Radar Imaging Based on ADMM-Net

Through-the-wall radar (TWR) images via the back projection (BP) imaging algorithm contain high levels of sidelobes, which makes it difficult to identify weak scattering targets such as the human body. The sparse reconstruction method based on compressed sensing (CS) can provide high-resolution TWR images with weak sidelobes. However, CS reconstruction requires considerable time for the iterative solving process and tuning of parameters. To solve the above problem, a deep unfolding network dubbed alternating direction method of multipliers net (ADMM-Net) is proposed for TWR imaging in the paper. The proposed imaging method maps the iterative subproblem solving process of the ADMM algorithm into a network, where an iterative process of the ADMM algorithm becomes one layer in the network topology. Each layer of the proposed ADMM-Net consists of three modules, i.e., the scattering coefficient reconstruction module, the nonlinear transformation module, and the multiplicative updating module. The experimental results demonstrate the superiority of the proposed TWR imaging method over the traditional BP and ADMM algorithms.