Modulated-ISRJ rejection using online dictionary learning for synthetic aperture radar imagery

In electromagnetic countermeasures circumstances,synthetic aperture radar(SAR)imagery usually suffers from severe quality degradation from modulated interrupt sampling repeater jamming(MISRJ),which usually owes considerable coherence with the SAR transmission waveform together with periodical modulation patterns.This paper develops an MISRJ suppression algorithm for SAR imagery with online dictionary learning.In the algorithm,the jamming modulation temporal properties are exploited with extracting and sorting MISRJ slices using fast-time autocorrelation.Online dictionary learning is followed to separate real signals from jamming slices.Under the learned representation,time-varying MISRJs are suppressed effectively.Both simulated and real-measured SAR data are also used to confirm advantages in suppressing time-varying MISRJs over traditional methods.

Simulation and Analysis of the Impacts of Anisotropic Ionospheric Scintallition on Geosynchronous Synthetic Aperture Radar

The impacts of ionospheric scintillation on geosynchronous synthetic aperture radar（GEO SAR）focusing is studied based on the multiple phase screen（MPS）theory.The power spectrum density of electron irregularities is first modified according to the ionospheric anisotropy.Then propagation wave equations in random medium are deduced in the case of oblique incidence in GEO SAR.The amplitude and phase errors induced by the random electron fluctuations are generated by the iterated MPS simulations and are superimposed into the GEO SAR signals.Through the following imaging and evaluation,the effects of the anisotropic ionospheric scintallition on GEO SAR are assessed.At last,the optimized integration time under different ionospheric scintillation conditions are recommended through Monte Carlo experiments.It is concluded that,greater ionospheric fluctuations and longer integration time will result in more severe deterioration,even no focus at all in the worst case.

Simulation of Raw Signal for Spaceborne Synthetic Aperture Radar(SAR)

The study on simulation of raw signal for spaceborne SAR aims at producing raw signal to test and evaluate the system and imaging algorithm. The model used for simulation includes a distributed target model, a platform and target geometry model, and a mathematical architecture used for generation of raw echo. Two aspects are stressed, one is the effects of earth ellip soid and attitude errors on radar impulse respense, the other is quick generation of range migration in azimuth frequency domain. Prescribed statistical characteristics of the model account for a realistic speckle of actual image. Finally, examples are given to validate the simulation of raw signal for spaceborne SAR.

An Information Theory Approach to the Data Compression and Imaging System for Synthetic Aperture Radar （SAR）

Synthetic aperture radar (SAR) is portrayed as a multiple access channel. An information theory approach is applied to the SAR imaging system, and the information content about a target that can be extracted from its radar image is evaluated by the average mutual information measure. A conditional (transition) probability density function (PDF) of the SAR imaging system is derived by analyzing the system and a closed form of the information content is found. It is shown that the information content obtained by the SAR imaging system from an independent sample of echoes will decrease and the total information content obtained by the SAR imaging system will increase with an increase in the number of looks. Because the total average mutual information is also used to define a measure of radiometric resolution for radar images, it is shown that the radiometric resolution of a radar image of terrain will be improved by spatial averaging. In addition, the imaging process and the data compression process for SAR are each treated as an independent generalized communication channel. The effects of data compression upon radiometric resolution for SAR are studied and some conclusions are obtained.

Carbon stock estimation by dual-polarized synthetic aperture radar (SAR) and forest inventory data in a Mediterranean forest landscape

Forest ecosystems play a crucial role in mitigating global climate change by forming massive carbon sinks. Their carbon stocks and stock changes need to be quantified for carbon budget balancing and international reporting schemes. However, direct sampling and biomass weighing may not always be possible for quantification studies conducted in large forests. In these cases, indirect methods that use forest inventory information combined with remote sensing data can be beneficial. Synthetic aperture radar (SAR) images offer numerous opportunities to researchers as freely distributed remote sensing data. This study aims to estimate the amount of total carbon stock (TCS) in forested lands of the Kizildag Forest Enterprise. To this end, the actual storage capacities of five carbon pools, i.e. above- and below-ground, deadwood, litter, and soil, were calculated using the indirect method based on ground measurements of 264 forest inventory plots. They were then associated with the backscattered values from Sentinel-1 and ALOS-2 PALSAR-2 data in a Geographical Information System (GIS). Finally, TCS was separately modelled and mapped. The best regression model was developed using the HH polarization of ALOS-2 PALSAR-2 with an adjusted R^(2) of 0.78 (p < 0.05). According to the model, the estimated TCS was about 2 Mt for the entire forest, with an average carbon storage of 133 t ha^(−1). The map showed that the distribution of TCS was heterogenic across the study area. Carbon hotspots were mostly composed of pure stands of Anatolian black pine and mixed, over-mature stands of Lebanese cedar and Taurus fir. It was concluded that the total carbon stocks of forest ecosystems could be estimated using appropriate SAR images at acceptable accuracy levels for forestry purposes. The use of additional ancillary data may provide more delicate and reliable estimations in the future. Given the implications of this study, the spatiotemporal dynamics of carbon can be effectively controlled by forest management when coupled with easily accessible space-borne radar data.

微波视觉与SAR图像智能解译

高分辨率雷达成像技术和人工智能、大数据技术的快速发展,有力促进了雷达图像智能解译技术的进步。由于雷达传感器本身的特殊性和电磁散射成像物理的复杂性,雷达图像的解译缺乏光学图像的直观性,准确迅速识别分类的需求对雷达图像解译提出了迫切的挑战。在借鉴人脑光视觉感知机理和计算机视觉图像处理相关技术基础上,进一步融合电磁散射物理规律及其雷达成像机理,我们提出发展微波域雷达图像解译的“微波视觉”的新交叉领域研究。该文介绍微波视觉的概念与内涵,提出微波视觉认知模型,阐述其基础理论问题与技术路线,最后介绍了作者团队在相关问题上的初步研究进展。

方位向调制干扰对高分宽幅多通道SAR的影响

多通道合成孔径雷达(synthetic aperture radar,SAR)能够利用方位向多通道重构处理去除多普勒模糊,获得高分宽幅SAR图像。针对此特殊的多通道重构处理方式,研究了传统SAR方位向调制干扰对高分宽幅多通道SAR的影响,包括多普勒调制干扰和脉冲延迟转发干扰两种类型的方位向调制干扰方式,为高分宽幅多通道SAR干扰决策提供理论基础。理论分析指出,多普勒调制干扰信号经成像处理后将在方位向生成等间隔分布的虚假目标,并且幅度受与多普勒调制频率相关的正弦函数和匹配滤波损失联合调制;而脉冲延迟转发干扰信号在成像后将只有单个干扰目标出现,且目标的方位位置仅与延迟的脉冲数有关。仿真实验验证了理论分析的正确性。

基于噪声卷积调制的SAR多区域灵巧干扰方法

针对传统噪声卷积干扰存在的压制区域位置在方位向上不能灵活控制的缺陷,提出一种基于噪声卷积调制的SAR多区域灵巧干扰方法。该方法对截获信号进行距离向移频、方位向移相以及与噪声模板的卷积调制处理,以控制压制区域的位置;后对截获信号进行不同频率、相位的噪声卷积调制处理,以控制压制区域的数量。理论和仿真结果表明,所提方法实现了压制区域位置在方位向、距离向上的灵活控制,并实现了单一干扰机对多个待掩护区域的干扰,提高了干扰增益,干扰效果较好,对实际工程应用具有一定的参考价值。

基于长短路融合及数据平衡的SAR船舶检测算法

针对SAR图像检测船舶任务中的目标小、近岸样本目标检测困难等问题,文章提出一种名为长短路特征融合网络(Long and Short path Feature Fusion Network,LSFF-Net)的船舶检测网络。该网络通过长短路特征融合模块有效协调了大目标与小目标检测,避免小目标特征信息的丢失。网络中应用结构重参数化结构提高了模块学习能力。为了满足多尺度目标检测,加入特征金字塔网络,融合多尺度特征。为了应对近岸样本目标检测,设计数据重分配算法,提高了对近岸样本目标的检测精度。实验结果表明:在公开数据集检测时,算法的平均精度(Average Precision,AP)达到97.50%,优于主流目标检测算法。该方法为提高SAR图像中小目标和近岸样本目标检测精度提供了新的实现方案。

多域特征引导的无监督SAR图像舰船检测方法

如何在合成孔径雷达(SAR)图像标注样本有限的条件下,提升舰船检测性能一直是SAR图像处理中的热点问题。本文提出一种多域特征引导的无监督域适应方法,将知识从有标注的源域(光学图像)转移到未标注的目标域(SAR图像),降低对标记SAR图像数据依赖。同时,设计了频域转换模块、注意力区域增强模块和自适应权重模块来缩小光学、SAR图像域之间的域差距,提高源域与目标域特征对齐效率,增强网络在挑战性样本下的特征迁移能力。在公开发布的数据集上进行了大量实验。结果表明:所提的模块较基础模型AP50提升10%,总体性能优于其他先进的方法。

基于遗传算法的无人机载穿墙三维SAR航迹规划方法

传统手持或车载穿墙雷达由于架设高度受限,无法对城市高层建筑内部目标进行透视成像,无人机载穿墙雷达具有灵活机动、高效便捷、无高度限制等优势,可对城市高层楼宇进行大范围三维穿透探测。三维层析合成孔径雷达(SAR)成像广泛采用多基线扫描模式,以获得高度向高分辨能力,但存在高度向空域欠采样导致的栅瓣问题。对此,该文提出一种基于遗传算法的无人机载穿墙三维SAR航迹规划方法,通过非均匀化飞行航迹,削弱周期性的雷达回波能量叠加,从而抑制栅瓣、实现更优的成像质量。该算法结合飞行距离与无人机载穿墙雷达成像质量的内在关系,建立了无人机航迹规划代价函数;利用遗传算法对3种典型的无人机飞行轨迹关键控制点进行基因编码,并进行基因杂交、变异等以优化种群与个体;最终通过最小化代价函数,分别筛选出3种典型飞行模式下的最优飞行航迹。仿真和实测数据的三维成像结果表明:相较于传统等间距多基线飞行模式,所提方法显著抑制了成像目标的栅瓣效应;此外,无人机斜线飞行的航迹长度明显缩短,提高了无人机载穿墙SAR成像效率。

考虑速度聚束效应的SAR海浪成像仿真方法

针对合成孔径雷达(SAR)海浪成像仿真技术对于速度聚束效应的考虑不充分,仿真数据无法准确反映实际SAR数据的问题,建立一种充分考虑速度聚束效应的SAR海浪成像仿真方法并进行了实验验证。计算了SAR图像强度的概率密度分布的仿真结果与理论结果之间的均方误差(MSE),在风速分别为5、10和15 m/s时,考虑速度聚束效应前、后比对结果的MSE分别为0.1012、0.1576、0.0556与0.0179、0.0314、0.0088。仿真结果表明,所提SAR海浪成像仿真方法有效提高了仿真数据的准确性,对SAR海浪成像的应用具有一定实用价值。

基于向量法近似的圆弧SAR成像方法

Arc SAR是一种能够在短时间内可重复扫描360°的雷达成像技术,具有重访周期短,观测范围广等优点,在边坡检测、山体滑坡等场景中有着广泛的应用前景。在Arc SAR模式下雷达随着机械臂进行圆周运动来形成合成孔径,因其运动轨迹不是直线,所以不能直接运用传统直线导轨的成像方法,需为Arc SAR设计新的成像算法,然而现有的成像方法不能很好地平衡精度和复杂度。在距离多普勒算法中,由于采用泰勒展开来近似斜距,导致成像质量不佳,或是需采取分割策略进行成像。因此,提出了一种新的距离多普勒算法,利用目标与雷达的向量关系进行近似,求得了近似斜距表达式,继而得到了距离多普勒域的信号表达式。基于这一表达式,提出了相应的相位误差补偿方法,完成了图像聚焦。最后通过点目标仿真验证了该算法的有效性。

GEO SAR照射的机载前视接收双基SAR系统二维分辨能力分析

地球同步轨道(Geostationary Earth Orbit,GEO)星/机双基前视合成孔径雷达(Synthetic Aperture Radar,SAR)实现了接收机飞行方向的成像能力,可以对重点区域进行持续前视观测,具有广泛的应用场景。对固定轨道GEO SAR照射下的机载前视接收机的双基SAR系统二维分辨能力进行了解析和验证仿真分析。首先,基于梯度法给出了GEO星/机双基前视SAR的二维分辨能力及其夹角的解析结果;其次,分析了不同双基前视构型对距离向分辨率、方位向分辨率和分辨率夹角的影响,并给出了基于分析结果的构型设计基本准则;最后,对GEO SAR处于不同轨道位置时分别进行了点目标仿真,有效验证了该分析方法的有效性和准确性。

基于GB-SAR的矿山边坡局部变形演化与失稳预测

为了提高地基合成孔径雷达(ground-based synthetic aperture radar, GB-SAR)边坡监测效果与失稳预测能力,在广东省某石灰石矿山架设地基雷达,开展实地监测试验,结合矿山地质条件与监测数据研究不同类型的岩体变形演化特征,根据变形特征选择位移最大点的监测数据利用速度倒数法进行失稳预测。通过分析典型案例的监测数据来优化速度倒数法的算法,进而对矿山试验数据进行研究。研究结果表明:通过分析雷达三维模型变形图的局部位移差异,能够初步判断边坡岩体的变形破坏方式;适当增大移动平均滤波法的移动时间周期C,能够更早地发现灾害征兆,更好地识别岩体加速变形阶段及临滑阶段。研究结果可为基于雷达的滑坡预测提供参考。

SAR图像飞机目标智能检测识别技术研究进展与展望

合成孔径雷达(SAR)采用相干成像机制,具有全天时、全天候成像的独特优势。飞机目标作为一种典型高价值目标,其检测与识别已成为SAR图像解译领域的研究热点。近年来,深度学习技术的引入,极大提升了SAR图像飞机目标检测与识别的性能。该文结合团队在SAR图像目标特别是飞机目标的检测与识别理论、算法及应用等方面的长期研究积累,对基于深度学习的SAR图像飞机目标检测与识别进行了全面回顾和综述,深入分析了SAR图像飞机目标特性及检测识别难点,总结了最新的研究进展以及不同方法的特点和应用场景,汇总整理了公开数据集及常用性能评估指标,最后,探讨了该领域研究面临的挑战和发展趋势。

GEO SAR天线波束指向定标中的误差分析

GEO SAR因其超宽测绘带和超长的合成孔径时间,导致低轨SAR波束指向定标方法不再适用,进而采取基于多脉冲分时比幅的波束指向定标方法。本文针对该指向定标方法,结合GEO SAR星地几何关系,对卫星姿态变化、三维系统性误差以及波束指向定标方法中接收机信噪比和通道增益稳定性引入的指向误差进行了详细的推导和仿真分析。仿真结果表明,0.003°的姿态测量误差会引入0.003°的距离向指向误差和0.0033°的方位向指向误差;三维系统性误差是导致天线波束指向变化的主要误差源;当SNR≥35 dB、通道增益稳定性优于1 dB时,GEO SAR波束指向定标方法引入的指向误差小于0.001°。

面向SAR图像目标分类的CNN模型可视化方法

卷积神经网络(CNN)在合成孔径雷达(SAR)图像目标分类任务中应用广泛。由于网络工作机理不透明,CNN模型难以满足高可靠性实际应用的要求。类激活映射方法常用于可视化CNN模型的决策区域,但现有方法主要基于通道级或空间级类激活权重,且在SAR图像数据集上的应用仍处于起步阶段。基于此,该文从神经元特征提取能力和网络决策依据两个层面出发,提出了一种面向SAR图像的CNN模型可视化方法。首先,基于神经元的激活值,对神经元在其感受野范围内的目标结构学习能力进行可视化,然后提出一种通道-空间混合的类激活映射方法,通过对SAR图像中的重要区域进行定位,为模型的决策过程提供依据。实验结果表明,该方法给出了模型在不同设置下的可解释性分析,有效拓展了卷积神经网络在SAR图像上的可视化应用。

大幅宽SAR图像嵌入式舰船实时检测系统设计

针对星载或机载高分辨率合成孔径雷达(synthetic aperture radar,SAR)实时成像后的大幅宽SAR图像舰船实时检测的应用需求,传统的基于FPGA+DSP的嵌入式系统很难同时实现SAR成像处理和基于人工智能技术的大幅宽SAR图像舰船实时检测,为此设计了一种基于3U VPX FPGA+GPU架构的大幅宽SAR图像嵌入式舰船实时检测系统;提出了一种基于YOLOv5s的舰船检测模型,采用基于L2-范数稀疏性惩罚的缩放因子控制法进行轻量化,轻量化舰船检测模型的参数量减小了47.39%,计算量减少了18.67%,平均检测精度为0.968;将轻量化舰船检测模型应用于大幅宽SAR图像嵌入式舰船实时检测系统,并针对典型的10 km×10 km的大幅宽图像应用场景,设计开发基于多线程技术和基于GPU的众核并行计算技术的大幅宽SAR图像嵌入式实时检测系统软件;通过公开的SAR数据集进行功能验证和性能评估,该系统能够满足不同分辨率的大幅宽SAR图像舰船实时检测需求。

基于多层显著性模型的SAR图像舰船目标检测

针对合成孔径雷达图像舰船目标检测问题,提出了一种结合选择机制与轮廓信息的多层显著性目标检测方法。首先,利用非下采样剪切波和频谱残差法进行全局显著性区域提取。其次,提出了一种基于动态恒虚警率的活动轮廓显著性模型,逐步滤除候选区域的虚警,提取目标轮廓,从而实现目标的精确检测。所提方法能够由粗到细地快速捕获目标区域,从而实现高效、高分辨率合成孔径雷达图像舰船检测。最后,在真实SAR数据集进行了测试,与其他经典的舰船检测方法相比,所提算法不仅有效地抑制了海杂波的影响,而且在检测精度上有较大提高。