基于SVD-SRNet的SAR三维成像方法

合成孔径雷达（SAR）三维成像是传统二维SAR成像在雷达精细信息获取与感知领域的重要发展，可分辨重叠于二维SAR图像同一像素中的多个目标。稀疏信号处理是进行SAR三维成像的有效方法，但由于稀疏信号处理的非线性特征，常需迭代运算，效率较低。研究人员已提出利用深度学习技术实现快速解算非线性信号处理问题的思路，在三维成像领域已有初步应用。然而，由于SAR三维实测数据稀缺，三维成像网络的训练只能依赖于仿真数据进行，并且仿真数据与实测数据存在差异大的问题，导致基于深度学习方法的SAR三维成像精度受限。为此，本文提出了一种基于奇异值分解的信号空间归一化超分辨网络（SVD Signal-Space Normalization Super-Resolution Net，SVD-SRNet），能够解决由于仿真数据与实测数据存在差异大导致的三维成像网络化方法鲁棒性低的问题，与传统方法相比所提方法具有更优异的成像精度。计算机仿真试验和无人机SAR实测数据试验证明了本文所提方法的有效性。 

SAR Three Dimensional Imaging Method Based on SVD-SRNet

? ?Three dimensional （3-D） Synthetic Aperture Radar （SAR） imaging is an important development of traditional two dimensional （2-D） SAR imaging in the field of radar precision information acquisition and perception， which can distinguish multiple targets overlapping in the same pixel of 2-D SAR images. Sparse signal processing is an effective method for 3-D SAR imaging. However， due to the nonlinear characteristics of sparse signal processing， it often needs iterative operation and has low efficiency. Researchers have proposed the idea of using deep learning technology to solve nonlinear signal processing problems quickly， which has been initially applied in 3-D imaging field. However， due to the scarcity of SAR 3-D real data， the training of 3-D imaging network can only rely on simulation data， and there is a difference between simulation data and real data， which results in limited accuracy of SAR 3-D imaging based on deep learning method. To this end， this paper proposes a signal space normalization super-resolution network based on singular value decomposition （SVD Signal-Space Normalization Super-Resolution Net， SVD-SRNet）. The proposed method can solve the problem of low robustness of 3-D imaging network method due to the large difference between simulation data and real data. Compared with traditional methods， the proposed method has better imaging accuracy. The computer simulation test and the UAV SAR measured data test prove the effectiveness of the method proposed in this paper.