Unsupervised Deep Basis Pursuit Based Resolution Enhancement for Forward Looking MIMO SAR Imaging

Nowadays, radar-based image reconstruction is becoming important in higher level automated driving, especially for all weather conditions. In this article, we present an unsupervised deep learning method for forward looking multiple-input multiple-output synthetic aperture radar (FL-MIMO SAR) to enhance the angular resolution. We present mathematical analysis for the composite antenna pattern generated by FL-MIMO SAR as well as image reconstruction with deep learning for FL-MIMO SAR. We present a computationally efficient deep basis pursuit (DBP) method to solve for convolutional neural network (CNN) with unsupervised learning (i.e., without ground truth) and present modified backprojection algorithm to reconstruct SAR image with enhanced angular resolution. We present experimental results to verify our proposed methodology and compare the performance with compressed sensing-based backprojection algorithm on both simulation and real data.