基于微波和毫米波的近场合成孔径雷达（SAR）压缩感知成像关键技术

To achieve the requirement of real-time or quasi-real-time imaging for the near-filed synthetic aperture radar (SAR) imaging system, this project attempts to employ the idea of compressed sensing into near-field SAR imaging, and focuses on the following problems: (1) key techniques of compressed sensing when applying to the Born linear imaging model, including the 2-D and 3-D near-field SAR imaging principle, optimal design of measurement matrix with low coherence; (2) key techniques of compressed sensing when applying to the Born nonlinear imaging model, including optimal feedforward neural network design based on iterative imaging, difference correction algorithm based on compressive inverse scattering method; (3) fast key compressed sensing algorithms: using iterative correction and nonlinear updating process to further reduce the computational complexity, using sparse noise projection method to improve the robustness; (4) optimal basis selection compressed sensing method, which is used to improve the recovered image quality with fixed low under-sampling rate. Besides, this project attempts to make a breakthrough in the following techniques: optimization design of 3-D measurement matrix based on truncation repair + sampling density compensation, iterative compressive imaging structure design based on feedforward neural network + difference correction, and optimal design sparsity basis method based on the improved selection. The research results are expected to achieve innovation on the three aspects: near-field SAR compressive imaging methods under weak and strong reflections, low computation complexity and high robustness algorithms and optimized sparsity basis selection algorithms.

为了满足实时或者准实时的近场合成孔径雷达（SAR）成像需求，本项目将压缩感知理论应用到近场SAR成像技术中，拟重点研究：1）Born线性成像模型下的压缩感知关键技术：近场SAR 2-D和3-D成像原理；低相干特性的测量矩阵优化设计；2）Born非线性成像模型下的压缩感知关键技术：基于迭代成像的前馈神经网络设计；基于压缩逆散射方法的差值校正算法；3）压缩感知的快速关键算法：采用迭代校正和非线性更新过程，减小算法的复杂度；利用稀疏的噪声投影方法，增强算法的鲁棒性；4）压缩感知的稀疏基选择优化方法，实现在较低欠采样率下提高被测物体的恢复图像质量。拟突破基于截断修复+采样密度补偿的3-D测量矩阵优化设计、基于前馈神经网络+差值校正的迭代压缩成像结构设计、不同优化准则下的稀疏基设计等关键问题。在面向弱反射和强反射情况下近场SAR压缩感知成像方法、低计算复杂度和高鲁棒性算法与稀疏基选择优化上取得创新。

近场毫米波合成孔径雷达（SAR）成像在穿墙成像、航空航天成像、无损检测与评价等民用、军事领域有着重要的应用，但是其长时间的数据采集过程严重限制了其应用范围和领域。. 为了满足实时或者准实时的成像需求，本项目研究了将压缩感知理论应用到近场毫米波成像中的几个关键内容：.（1）设计、 加工并测试完成面向近场成像应用的极小型天线，并构建Ka-Band 30GHz~40GHz 的近场成像系统。.（2）近场 SAR-2D 压缩感知矩阵优化设计及快速算法，将高精度成像采样率下降到10%。.（3）近场 SAR-3D 压缩感知矩阵优化设计及快速算法，设计了基于截断修复+采样密度补偿的SAR-3D前向/后向快速变换对，使得正交SAR-3D前向/后向快速变换对的误差到 10^-15量级；以成像系统PSF为优化方向，设计了确定性低采样率下的采样矩阵优化设计方法，将高精度成像采样率下降到6%~7%。.（4）非线性近场SAR超分辨图像增强算法，将近场SAR超分辨图像增强问题转化为原始-对偶框架下的优化最小化问题，从稀疏观测数据中直接重建高分辨图像。.（5）考虑实际噪声情况下的近场SAR压缩感知成像的快速关键算法，从半二次方HQ和广义近似消息AMP框架下，赋予近场毫米波成像算法对非规则特性噪声的成像鲁棒性。.（6）为了减轻算法参数调整的复杂度，研究近场SAR的核自适应滤波算法。根据其和压缩感知恢复问题的相似性，提出了非线性情况下的基于Student’s-t的核自适应滤波方法和基于L1范数的核自适应滤波方法，降低神经网络的实现复杂度和减少算法迭代次数。. 本项目研究成果有助于实现近场毫米波实时成像，为实现我国近场毫米波成像技术的应用推广，发挥一定作用。