【作者】 刘记红；

【导师】 庄钊文； 黎湘；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2012， 博士

【摘要】 高分辨ISAR成像技术对于雷达目标识别和特征提取具有重要意义。本文针对传统ISAR成像系统面临的采样率高、数据量大、回波数据有限及目标复杂运动条件下方位向采样不均匀时成像困难等问题，充分利用雷达目标散射率分布的稀疏性和压缩感知信息处理的巨大优势，围绕基于压缩感知的ISAR成像技术展开研究，重点研究了基于压缩感知的转台目标成像技术、高速运动目标成像技术和复杂运动目标成像技术。第一章阐述了课题研究背景及意义，介绍了高分辨成像雷达的发展概况和ISAR成像技术研究现状，概述了压缩感知理论的发展及其应用，对基于压缩感知的雷达成像技术进行了归纳总结和分析，最后介绍了本文的主要研究工作。第二章为基础理论研究。首先介绍了压缩感知的基本原理，对压缩感知的数学模型及关键要素进行了分析和讨论；然后从光学区散射中心理论、雷达成像根本原理的角度分析了雷达回波数据的稀疏性机理；最后从系统实际出发，提出了基于随机卷积的压缩感知雷达成像方法，该方法能够在少量观测数据条件下获得较好的成像结果，便于物理实现，且不受雷达发射波形的限制。第三章研究了基于压缩感知的转台目标成像技术和快速重构算法。针对宽带线性调频雷达直接采样面临的数据采集和存储压力，构造了一种包含Stretch处理和傅里叶变换信息的稀疏字典，据此提出一种基于压缩感知的成像算法，所提方法省略了解线频调步骤，在实现高分辨成像的同时大大降低了雷达成像系统的数据率。针对频率步进雷达数据利用率低的问题，提出了基于压缩感知的二维联合和二维解耦成像算法。两种算法均可利用少量测量数据获得清晰的ISAR图像，并由于将相干混频处理过程融入稀疏字典中，简化了雷达系统的硬件设计。针对压缩感知成像算法复杂度高的问题，根据稀疏字典及测量矩阵的二维可分离特性，研究了压缩感知成像的快速重构算法，并提出一种改进的贪婪算法用于雷达图像重构，大大降低了存储量和计算量，提高了成像效率。第四章研究了不同雷达信号体制下基于压缩感知的高速运动目标成像技术。针对线性调频雷达，根据解调后高速目标回波在分数阶傅里叶域的稀疏性，提出采用模拟信息转换方式对回波进行压缩测量，通过非线性优化重构雷达目标图像，并以重构一维距离像的稀疏性为准则，采用黄金分割法搜索最佳变换阶数来确定稀疏字典。所提方法无需额外的速度补偿步骤，同时解决了成像模糊和宽带雷达数据量过大的问题。针对频率步进雷达数据利用率低，且敏感于多普勒的问题，利用相位对消技术和脉冲重复间隔设计原理，提出一种基于随机频率步进波形设计的压缩感知成像方法，能够在降低数据率的同时克服多普勒效应的影响。针对线性调频步进雷达总数据率较高，且脉间压缩敏感于多普勒的问题，提出了基于随机调频步进波形设计的压缩感知成像方法，能够在降低总数据率的同时，获得运动目标的高分辨距离像。实验结果表明，对于高速运动目标，利用所提CS成像方法能够获得高质量的ISAR图像。第五章研究了基于压缩感知的非匀速旋转目标、高速自旋目标和弹道中段进动目标成像技术。针对稀疏孔径和短孔径条件下非匀速旋转目标的ISAR成像问题，根据运动补偿后目标回波在匹配傅里叶域的稀疏性，提出了基于压缩感知的成像方法，并基于稀疏表示和优化搜索实现相对旋转参数的估计。所提方法解决了有限脉冲数据与方位分辨率之间的矛盾，且成像效果优于现有方法。针对高速自旋目标的二维/三维成像问题，提出了基于轨道运动的二维成像方法，利用压缩感知思想有效降低了所需采集的脉冲数。在此基础上利用自旋信息通过后向投影变换或压缩感知方法进行二维投影切片成像，从而得到目标三维散射点的相对位置分布。该方法通过预先获得散射点的高度维信息，大大降低了三维成像的复杂度。针对弹道中段进动目标的ISAR成像问题，通过对旋转对称目标回波模型的线性化处理，引入压缩感知思想，基于少量回波数据实现了进动目标的高分辨成像。所提方法相比现有方法改善了成像质量，提高了对进动参数的稳定性。第七章总结了论文的研究工作和主要创新点，指出需要进一步研究的问题。更多还原

【Abstract】 High resolution inverse synthetic aperture radar (ISAR) imaging technique is of greatsignificance to radar target recognition and feature extraction. Taking full advantage ofthe sparsity of radar target reflectivity and the direct information sampling property ofcompressed sensing (CS), this dissertation focuses on ISAR imaging technique based onCS, serving for overcoming the inherent limitations of traditional ISAR imagingsystems. The main research efforts include the high resolution imaging techniques ofrotary platform targets, high-speed targets and targets with complex motion.Chapter1illustrates the background and significance of this research subject, andintroduces the present status of high resolution imaging radar and ISAR imagingtechnique. Then the development and applications of CS theory, particularly in radarimaging domain, are reviewed and summarized, after which the main content of thisthesis is presented briefly.Chapter2introduces the basic principle of CS, and makes a careful analysis of themathematic model and essentials of CS firstly. Then, the sparsity mechanism of radarechoed signal is analyzed based on the scattering center theory in high frequencydomain and radar imaging principle. Finally, an imaging method based on CS byrandom convolution is studied from a practical perspective, which is convenient torealize and applicable to various radar waveforms.Chapter3focuses on the CS-based imaging technique of rotary platform targets andfast reconstruction algorithms. Firstly, to alleviate the direct sampling pressure ofwideband linear frequency modulated (LFM) radar, a sparse dictionary constructionmethod based on stretch processing and FFT operation is presented, following which anovel compressive radar imaging method is proposed, and it can remarkably reduce thedata rate of radar imaging system while maintaining imaging quality. Then, in view ofthe low data usage factor of stepped frequency radar, two imaging methods, termedCS-based2D joint imaging algorithm and CS-based2D decoupled imaging algorithm,are proposed. Both can get clear ISAR image with less data samples, and simplify thehardware design of radar system because the coherent mixing operation is incorporatedinto the sparse dictionary. Afterwards, considering the high computation complexity ofCS imaging methods, in light of the2D separability of sparse dictionary andcompressive measurement, several fast algorithms for radar image formation are studied,and an improved OMP algorithm is proposed, which possesses prominent superiorityover conventional CS algorithms in terms of storage and computation.Chapter4researches into the CS-based imaging technique of high-speed targets. ForLFM radar, according to the sparsity of the dechirped high-speed target echo infractional Fourier domain, Analog-to-Information Conversion (AIC) is recommended totake compressive measurements, following which the radar image can be recovered via nonlinear optimization, and the efficient golden section method is adopted to find theoptimal transform order taking the sparsity of range profile as search criteria. Theproposed method can achieve high resolution imaging of high-speed targets withoutadditional velocity compensation, but with less data samples and higher imaging quality.In allusion to the sensitivity to Doppler and low data usage factor of stepped frequencyradar, making use of the predesigned pulse repetition interval (PRI) and/or the phasecancellation principle, a compressive imaging method based on random steppedfrequency waveform design is put forward, which can effectively reduce the data ratewhile overcoming the negative influence of Doppler effect. Similarly, as for linearlymodulated stepped frequency (LMSF) radar, to overcome the sensitivity of inter-pulsecompression to Doppler and reduce the total data rate, a compressive imaging methodbased on random LMSF waveform design is proposed. Simulation experiments indicatethat the proposed CS-based methods can get high quality ISAR images of targets withhigh-speed motion.Chapter5studies the CS-based imaging technique of targets with complex motion.First, in light of the sparsity of the range cell echo of radar targets with non-uniformrotation in matching Fourier domain, an imaging method based on CS is proposed forboth sparse aperture and short aperture cases, and the relative rotation parameter isestimated by the optimal search in fractional Fourier domain for the sparse dictionaryconstruction. The proposed method resolves the contradiction between limited pulsedata and high azimuth resolution, and performs better than the existing methods. Second,for rapidly spinning targets, a2D imaging method using the orbital information ispresented, where CS is incorporated for reducing the requisite pulses. Then, thespinning information is utilized for2D slice imaging via the complex-valuedback-projection transform or CS method, consequently the high resolution3D image isobtained. The method proposed here reduces the complexity of3D imaging effectivelythrough pre-estimating the altitude information of scattering centers. Third, CS conceptis introduced into the ISAR imaging problem of precession targets with rotationallysymmetrical structure, and high resolution imaging can be achieved with a smallnumber of measurements. The proposed method performs better than existing methodsin terms of imaging quality and robustness to precession parameter.Chapter6summarizes the research work and main innovations, and points out thefuture work to be researched.更多还原