Imaging for Forward Looking MIMO SAR with Un-Trained Neural Network

In the recent years, un-trained convolutional neural networks (CNN) have achieved excellent performance for image reconstruction problems, in the absence of training data. In this paper, we adopt an un-trained neural network (namely, Convolutional decoder) for forward looking multiple-input multiple-output synthetic aperture radar (FL-MIMO-SAR) to improve the angular resolution, followed by modified back projection (MBP) algorithm to reconstruct the final estimate of the FL-MIMO-SAR image. We show that our proposed method performs well especially in the case of low number of available measurements. We present simulation results to verify our proposed methodology, and compare the performance with deep basis pursuit (DBP) based back projection algorithm.