基于循环神经网络和字典学习的无源多基地SAR稀疏观测成像技术研究

The focal quality of imagery derived via coherent processing of passive multistatic SAR data will degrade significantly when there are gaps between the observed angles of different receivers, phase errors and azimuth weighting introduced by flight deviation and anisotropic scattering, respectively. To solve the above problems, this program will make research on the ultra-high resolution imaging and refocusing algorithm for the anisotropic scattering targets for passive multistatic SAR data with sparse observations to implement accurate focusing of the imagery. Firstly, an unsupervised recurrent neural network machine learning method is proposed to estimate the parameters of filters and realize accurate focusing of the received signal. Then, an super-resolution methodology based on joint dictionary learning is presented to acquire the resolution derived with 360°observation angle. Finally, the refocusing algorithm based on sparse deconvolution is designed for the imagery of anisotropic scattering targets. The aim of this program is to obtain a high-resolution SAR imagery based on passive multi-static SAR with sparse observations, which can be viewed as a theory and fundamental supplement for the development of passive SAR.

无源多基地合成孔径雷达(Synthetic Aperture Radar,SAR)数据采集条件下，不同平台观测角度之间可能存在缺失，航迹偏移以及各向异性电磁散射又会在回波中分别引入相位误差和方位向加权，导致直接相干处理重建图像质量严重下降。为解决此问题，本项目拟研究构建无源多基地SAR稀疏观测信号的超高分辨成像算法以及各向异性电磁散射区域重聚焦方法，得到高精度聚焦图像。首先，在航迹偏移未知的条件下，利用循环神经网络无监督学习方法实现滤波器参数的估计以及高精度聚焦成像。然后，基于多平台接收观测数据，设计联合字典学习方法重建图像获得全方位观测分辨性能。最后，对各向异性散射目标图像构建了基于稀疏盲解卷积的重聚焦处理方案。本项目研究旨在利用无源多基地SAR部分观测角度数据，获得具备全方位观测分辨性能的高精度聚焦图像，从而为未来我国无源高分系统的研制提供理论储备和技术支持。

无源多基地合成孔径雷达(Synthetic Aperture Radar,SAR)采集的回波中可能存在角度缺失，航迹偏移以及各向异性电磁散射，导致重建图像质量下降。为解决此问题，本项目研究了无源多基地SAR稀疏观测信号的超高分辨成像算法以及各向异性电磁散射区域重聚焦方法。首先，在非理想航迹条件下，结合相位误差估计和循环神经网络无监督学习方法实现了对单平台采集回波的匹配滤波器参数估计以及高精度聚焦成像。然后，对多平台接收的稀疏观测数据，设计联合字典学习方法提高重建图像的分辨率。最后，对各向异性散射目标回波设计了基于联合稀疏约束的信号分解和成像方案。本项目研究成果实现了无源多基地SAR部分观测角度数据的超分辨率成像处理并提高了各向异性散射目标图像的可解译性，从而为未来我国无源高分系统的研制提供理论储备和技术支持。