基于TMS320C6701高速并行信号处理平台的设计

80年代初，数字信号处理（DSP）在合成孔径雷达（Synthetic Aperture Radar，简称SAR）实时成像处理中得到了应用，随着SAR技术的发展，对实时成像处理器的数据处理能力提出更高要求。本文介绍了如何设计出高速并行信号处理平台来满足SAR系统对实时成像处理器的要求，并给出基于该平台实现合成孔径雷达距离向脉冲压缩的实例。     

地球单基ISAR对月球成像信号参数设计与仿真

利用地基逆合成孔径雷达（Inverse Synthetic Aperture Radar,ISAR）对月球进行观测和成像处理,可获得月球正面的地形地貌等特征。本文首先构建了用于逆合成孔径雷达成像的地面单基雷达对月球观测的等效转台模型,分析了月球表面等效散射中心与雷达径向距离的变化特性。然后,给出了单基雷达脉冲重复频率、脉冲宽度、工作频段、带宽和采样频率等信号参数的设计准则,并建立了单基雷达对月球观测的ISAR回波模型与成像处理流程。最后,通过仿真实验验证了本文理论分析和成像算法的有效性。     

低轨卫星目标干涉ISAR三维成像方法

研究了在低轨道上运行的卫星目标的逆合成孔径雷达（Inverse Synthetic Aperture Radar，ISAR）成像问题，提出将干涉逆合成孔径雷达（Interferometric ISAR，IF-ISAR）成像技术应用于对卫星目标成像，并用此方法实现了对低轨卫星的三维成像。研究设计了卫星轨道模型和雷达天线空间配置，建立了卫星目标三维成像的几何坐标系，并对干涉成像的三维分辨率进行了估计。仿真结果验证了理论分析的正确性和成像方法的有效性。     

利用CUDA实现的基于GPU的SAR成像算法

高速发展的图形处理器（Graphics Processing Unit,GPU）为高效合成孔径雷达（Synthetic Aperture Radar,SAR）成像算法提供了具有发展前景的新型运算平台。与CPU相比,利用GPU进行通用计算具有成本低、性能高的特点。提出利用CUDA实现的基于GPU的SAR成像算法,与传统的基于CPU的成像算法相比,有两位数以上的效率提升,为应对SAR信号处理领域新的挑战提供具有前景的研究方向。     

合成孔径雷达干扰技术的研究

合成孔径雷达（Synthetic Aperture Radar）是一种新型高辨力成像雷达。其在距离上利用发射的线性调频（LFM）的脉冲压缩信号来获取距离上的高分辨力，在方位上利用接收脉冲之间的多普勒信息获得方位上的高分辨力。本文在简单分析合成孔径雷达工作原理基础上，针对合成孔径雷达的特点与弱点，讨论了几种干扰样式。     

一种正前视宽扇区高分辨雷达扫描成像方法

合成孔径雷达（Synthetic Aperture Radar,SAR）只有照射到飞行方向左、右两侧的感兴趣成像区域时才能处理出高分辨率SAR图像,飞行方向正前方的成像区域就成为了SAR成像的固有盲区。三通道（和、方位差、俯仰差）毫米波单脉冲成像雷达能够实现对正前视场景的二维成像,获取单脉冲图像。提出对SAR/单脉冲图像进行图像配准和融合拼接的方法：首先利用多尺度多方向二维Gabor滤波器组分别对SAR/单脉冲图像进行特征提取,然后对两组特征矩阵进行归一化互相关匹配,对匹配好的图像进行像素级融合处理,得到完整的正前视宽扇区高分辨雷达图像。试验数据成像处理结果表明,所提复合成像算法能够对飞行正前方宽扇区范围内进行高分辨成像,有效解决了工程实际中碰到的正前视高分辨成像盲区的难题,对于前视雷达成像侦察、弹载雷达目标区域景象匹配、飞行器夜航、盲降等具有一定的工程实际意义。     

MIMO雷达三维干涉诊断成像方法

2维合成孔径雷达(Synthetic Aperture Radar, SAR)和逆合成孔径雷达(Inverse Synthetic Aperture Radar, ISAR)成像是目标散射机理高分辨率分析和散射诊断测量的重要手段,现有被广泛采用的技术主要包括转台ISAR和导轨SAR成像技术。相比于传统的2维成像,3维成像可以提供目标局部散射中心在空间的3维位置和散射强度信息。因此,探索新的、可工程化实现与应用的目标3维成像技术是一项极具吸引力的工作。该文提出一种基于多输入多输出(Multiple-Input Multiple-Output, MIMO)阵列技术的3维干涉成像方法。首先,设计并试验了一种具有高孔径利用率和通过虚拟孔径实现干涉成像功能的MIMO阵列;其次,分析了MIMO阵列合成的两组虚拟孔径所成两幅2维雷达像的干涉相位与目标散射中心高度之间的关系,发展了MIMO雷达3维干涉成像算法;最后,通过数值仿真和原理样机实验验证了所提方法在目标散射机理分析和诊断测量应用中的可行性和有效性。      

基于对抗攻击的SAR舰船识别卷积神经网络鲁棒性研究

随着深度学习技术的迅猛发展,卷积神经网络(Convolutional Neural Networks,CNN)在合成孔径雷达(Synthetic Aperture Radar,SAR)舰船分类任务上取得很高的精度。但同时,由于SAR成像存在相干斑噪声等特性以及CNN自身的脆弱性,使得预测结果稳定性较差,在实际应用中存在明显隐患。针对上述CNN在SAR舰船识别分类任务上鲁棒性不足的问题,本文将对抗样本引入到SAR舰船识别鲁棒性的研究之中,通过从梯度、边界、黑盒模拟等多个角度对CNN网络进行全方位的对抗攻击及干扰,实现了对各SAR舰船识别CNN网络的综合评估,并依照评估结果完成针对性的鲁棒性增强方案的制定,为SAR舰船识别鲁棒性研究开拓了新的领域。      

基于全卷积神经网络的SAR图像目标分类

近年来,卷积神经网络(Convolutional Neural Network,CNN)在合成孔径雷达(Synthetic Aperture Radar,SAR)图像目标分类中取得了较好的分类结果。CNN结构中,前面若干层由交替的卷积层、池化层堆叠而成,后面若干层为全连接层。全卷积神经网络（All Convolutional Neural Network, A-CNN）是对CNN结构的一种改进,其中池化层和全连接层都用卷积层代替,该结构已在计算机视觉领域被应用。针对公布的MSTAR数据集,提出了基于A-CNN的SAR图像目标分类方法,并与基于CNN的SAR图像分类方法进行对比。实验结果表明,基于A-CNN的SAR图像目标分类正确率要高于基于CNN的分类正确率。     

层析SAR技术研究进展

合成孔径雷达(Synthetic Aperture Radar,SAR)通过将三维场景投影到距离-方位平面获取场景二维成像信息,但由此导致的叠掩问题影响图像解译。层析合成孔径雷达(Tomographic Synthetic Aperture Radar, TomoSAR)利用同一场景的多次航过数据在高程向合成孔径,获取目标三维成像结果,能够从根本上解决叠掩问题。本文从层析SAR信号模型出发,描述了不同维度下的层析SAR成像算法,系统地总结了应用于层析SAR研究的各机载和星载系统,阐述了层析SAR在城区、森林和冰川等场景下的应用,对于城区建筑物信息提取、森林参数信息提取和冰川结构参数提取中面临的主要问题和现有解决策略进行了详细的分析。最后,总结本文内容并展望层析SAR技术的发展前景。     

基于深度学习的复杂沙漠背景SAR目标检测

SAR目标检测,因成像场景大、背景复杂多变而极具挑战。传统基于恒虚警率的SAR目标检测方法极易受背景干扰。针对上述问题,提出一种基于深度学习的复杂沙漠背景SAR目标端对端检测识别系统。即采用小规模沙漠背景下的SAR图像数据对Faster-RCNN网络进行迁移训练,一体化完成典型目标的检测与识别。基于合成数据集Desert-SAR的试验结果表明,与传统方法相比,该方法检测速度更快、准确率更高、鲁棒性更强。     

基于深度学习的SAR图像舰船目标检测综述

近年来,合成孔径雷达成像技术因具备全天时和全天候的目标感测能力,在海洋实时监测和管控等领域发挥着重要作用,特别是高分率SAR图像中的舰船目标检测成为当前的研究热点之一。首先分析基于深度学习的SAR图像舰船目标检测流程,并对样本训练数据集的构建、目标特征的提取和目标框选的设计等关键步骤进行归纳总结。然后对检测流程中的各部分对SAR图像舰船目标检测精度和速度的影响进行对比分析。最后根据当前研究现状,深入分析深度学习算法在舰船检测应用中存在的问题,探讨基于深度学习的SAR图像舰船目标检测的进一步研究方向。     

凝视数字多波束合成孔径雷达原理

针对传统合成孔径雷达在实现高分辨大测绘带(简称高分宽幅)成像中存在的问题,文中将数字波束形成技术和合成孔径雷达结合,提出凝视数字多波束合成孔径雷达的概念。首先,将成像区域在距离和方位向分割为多个子区域进行独立成像,避免传统合成孔径雷达在高分宽幅成像领域的设计限制;然后,利用数字波束形成技术对天线阵面所有接收单元进行数字加权,在方位向和俯仰向形成多个高增益的接收窄波束,并控制接收波束实现对每个子区域的凝视观测,延长合成孔径的长度,提高成像分辨率。仿真表明,凝视数字多波束合成孔径雷达可以有效突破传统合成孔径雷达系统设计上的瓶颈,实现高分宽幅成像。     

A system model and inversion for synthetic aperture radar imaging

A system model and its corresponding inversion for synthetic aperture radar (SAR) imaging are presented. The system model incorporates the spherical nature of a radar's radiation pattern at far field. The inverse method based on this model performs a spatial Fourier transform (Doppler processing) on the recorded signals with respect to the available coordinates of a translational radar (SAR) or target (inverse SAR). It is shown that the transformed data provide samples of the spatial Fourier transform of the target's reflectivity function. The inverse method can be modified to incorporate deviations of the radar's motion from its prescribed straight line path. The effects of finite aperture on resolution, reconstruction, and sampling constraints for the imaging problem are discussed.     

Ocean wave extraction from RADARSAT synthetic aperture radarinter-look image cross-spectra

This study is concerned with the extraction of directional ocean wave spectra from synthetic aperture radar (SAR) image spectra. The statistical estimation problem underlying the wave-SAR inverse problem is examined in detail in order to properly quantify the wave information content of SAR. As a concrete focus, a data set is considered comprising six RADARSAT SAR images co-located with a directional wave buoy off the east coast of Canada. These SAR data are transformed into inter-look image cross-spectra based on two looks at the same ocean scene separated by 0.4 s. The general problem of wave extraction from SAR is cast in terms of a statistical estimation problem that includes the observed SAR spectra, the wave-SAR transform, and prior spectral wave information. The central role of the weighting functions (inverse of the error covariances) is demonstrated, as well as the consequence of approximate (based on the quasilinear wave-SAR transform) versus exact linearizations on the convergence properties of the algorithm. Error estimates are derived and discussed. This statistical framework is applied to the extraction of spectral wave information from observed RADARSAT SAR image cross-spectra. A modified wave-SAR transform is used to account for case-specific geophysical and imaging effects. Analysis of the residual error of simulated and observed SAR spectra motivates a canonical form for the SAR observation error covariance. Wave estimates are then extracted from the SAR spectra, including wavenumber dependent error estimates and explicit identification of spectral null spaces where the SAR contains no wave information. Band-limited SAR wave information is also combined with prior (buoy) spectral wave estimates through parameterization of the wave spectral shape and use of regularization     

Blind-velocity SAR/ISAR imaging of a moving target in a stationarybackground

A synthetic aperture radio/inverse synthetic aperture radar (SAR/ISAR) coherent system model and inversion to image a target moving with an unknown constant velocity in a stationary background are presented. The approach is based on a recently developed system modelling and inversion principle for SAR/ISAR imaging that utilizes the spatial Fourier decomposition of SAR data in the synthetic aperture domain to convert the SAR system model's nonlinear phase functions into linear phase functions suitable for a computationally manageable inversion. It is shown that SAR/ISAR imaging of a moving target can be converted into imaging the target in a stationary squint-mode SAR problem where the parameters of the squint-mode geometry depend on the target's velocity. A method for estimating the moving target's velocity that utilizes a spatial Doppler analysis of the SAR data within overlapping subapertures is presented. The spatial Doppler technique does not require the radar signal to be narrowband, so the reconstructed image's resolution is not sacrificed to improve the target's velocity estimator.     

基于SOMP的机载无源宽角SAR成像算法

针对机载无源宽角合成孔径雷达(SAR)在探测过程目标散射特性随观测角度变化的问题,在建立机载无源宽角SAR信号模型的基础上将成像重建过程转换为求解多测量向量的联合稀疏优化问题,并在分布式贪婪稀疏重建方法框架下提出一种基于同时正交匹配追踪的机载无源宽角SAR稀疏成像算法,能够实现对地面探测目标的准确稳健成像。仿真数据处理结果验证了所提成像算法的有效性和准确性。     

FM—CW SAR系统及其发展概况

FM-CW SAR是一种新近被提出来的成像雷达体制，它结合连续波与合成孔径成像技术，具有体积小、重量轻、成本低、分辨率高等一系列优点。文章概述了FM-CW SAR系统及其信号处理特点，讨论了FM-CW SAR系统实现的难题所在；归纳了目前FM-CW SAR的几种应用背景及其主要参数；根据FM-CW SAR发展概况，讨论了FM-CW SAR进一步的发展趋势。     

Derivation and comparison of SAR and frequency-wavenumber migration within a common inverse scalar wave problem formulation

Two common Fourier imaging algorithms used in ground penetrating radar (GPR), synthetic aperture radar (SAR), and frequency-wavenumber (F-K) migration, are reviewed and compared from a theoretical perspective. The two algorithms, while arising from seemingly different physical models: a point-scatterer model for SAR and the exploding source model for F-K migration, result in similar imaging equations. Both algorithms are derived from an integral equation formulation of the inverse scalar wave problem, which allows a clear understanding of the approximations being made in each algorithm and allows a direct comparison. This derivation brings out the similarities of the two techniques which are hidden by the traditional formulations based on physical scattering models. The comparison shows that the approximations required to derive each technique from the integral equation formulation of the inverse problem are nearly identical, and hence the two imaging algorithms and physical models are making similar assumptions about the solution to the inverse problem, thus clarifying why the imaging equations are so similar. Sample images of landmine-like targets buried in sand are obtained from experimental GPR data using both algorithms.     

A tomographic formulation of spotlight-mode synthetic aperture radar

Spotlight-mode synthetic aperture radar (spotlight-mode SAR) synthesizes high-resolution terrain maps using data gathered from multiple observation angles. This paper shows that spotlight-mode SAR can be interpreted as a tomographic reeonstrution problem and analyzed using the projection-slice theorem from computer-aided tomograpy (CAT). The signal recorded at each SAR transmission point is modeled as a portion of the Fourier transform of a central projection of the imaged ground area. Reconstruction of a SAR image may then be accomplished using algorithms from CAT. This model permits a simple understanding of SAR imaging, not based on Doppler shifts. Resolution, sampling rates, waveform curvature, the Doppler effect, and other issues are also discussed within the context of this interpretation of SAR.