Analysis of Deep Learning 3-D Imaging Methods Based on UAV SAR

As an important development of traditional SAR 2-D imaging, Synthetic aperture radar (SAR) 3-D imaging's core is sparse signal processing. However, due to the nonlinear characteristics of sparse signal processing, it often needs iterative calculation, which makes it inefficient. Researchers have put forward some ideas of using deep learning neural networks to quickly solve nonlinear signal processing problems, but it is lack of comparative analysis of different network performances. Therefore, this paper analyzes the abilities of two deep learning neural networks (ISTA-Net and ADMM-Net) to solve the 3-D imaging problem of tomographic SAR. Their quantitative performance in imaging accuracy and imaging efficiency is emphatically discussed, which can provide theoretical reference for subsequent deep learning SAR 3-D imaging research. The effectiveness of the analysis is verified by the measured data of UAV SAR.