Joint Matched Filtering and Iterative Optimization Network for 3-D mmW Imaging

Abstract

Compressed sensing (CS) shows significant potential to improve image quality in 3-D millimeter wave (mmW) imaging. Limited by huge computational cost, it always hard to apply conventional linear measurement-model-based imaging method for extended-scene reconstruction. In this paper, we present a joint Matched Filtering (MF) and Iterative Optimization network for 3-D mmW imaging, which is dubbed as MFIST-Net. First, the Matched Filtering kernels are introduced in IST optimization steps in lieu of measurement matrices, by which the large-scale matrix-vector operations in conventional linear measurement model are avoided and the computational efficiency is improved. Second, the modified IST is unfolded into a deep iterative architecture, and the parameters of MFIST-Net is learned from 1000 simulated data samples by end-to-end training. The well-trained model is capable to produce large-scale 3-D images of the illuminated targets from sparsely sampled echoes. Besides, numerical and visual results validate the proposed MFIST-Net achieve both favorable reconstruction accuracy and high execution time compared with MF, RMA, and ISTA algorithms.