机载SAR快速后向投影成像算法研究

发布时间：2018-09-08 13:29

【摘要】：合成孔径雷达(SAR)成像技术具有全天候、全天时、远距离作用、高分辨、广域观测的特点,能有效地提高了雷达的信息获取能力,因而被广泛地用于土地监测、海洋观测、海冰监视和地貌变化等民事领域以及战场侦察、军队动向监视等军事领域。随着SAR技术的发展,SAR正在朝着波束指向更灵活、分辨率更高、场景覆盖面积更大的方向发展,因此精确地获取地表信息具有重要意义。由于波束指向的灵活性,SAR具备多种工作模式,包括聚束模式、条带模式、滑动聚束模式和TOPS模式等。同时,每种工作模式又可以衍生出正侧视和斜视两种几何构型。在不同的工作模式、不同的几何构型下,数据录取方式及回波信号的形式差异较大,需要研究新的成像算法以适应不同工作模式的需求。本文针对机载SAR多模式成像及运动补偿中存在的关键问题,围绕国家自然科学基金、高等学校博士学科点专项科研基金、民口973项目开展研究,旨在拓展快速时域算法的适用范围和改善算法的运算效率。论文的主要工作如下：第二章是本文的基础理论章节,介绍了目前几种典型的SAR成像算法,讨论了每种算法的成像原理、关键技术以及优势和不足。根据波数支撑区的特点,第二章首先介绍了针对斜视聚束SAR极坐标处理的两种重采样方法,即沿视线极坐标插值(LOSPI)和固定场景插值(SSPI),推导了重采样之后的二维波数谱。需要指出的是,PFA不但生命力顽强、广泛用于高分辨聚束SAR、CSAR和视频SAR等,而且距离-多普勒域聚焦的成像特点和LOSPI为第三章开展基于快速时域算法的自聚焦处理提供了借鉴意义。此外,第二章还重点介绍了BP算法、FBP算法和FFBP算法三种时域算法,讨论了图像质量对真实的APC位置和地形起伏的依赖性。在实际应用中,大部分运动误差可根据GPS/INS记录的平台运动信息得到校正。然而在高分辨SAR成像场合,即使几厘米量级的残余运动误差都极可能对图像聚焦产生影响,因此快速时域算法有必要开展基于图像或数据的自聚焦处理。为此,第三章对FFBP算法进行了必要的改进。第一、选用LOS虚拟极坐标网格代替原始FFBP算法中的极坐标网格。使用该坐标系重建的图像不但可以提供图像域与距离压缩相位历程域之间的傅里叶变换关系,而且能够获得沿水平方向散布的脉冲响应函数。第二、利用FFBP算法的多孔径结构建立重叠子孔径构型(OSF)。 OSF是连接子孔径相位误差和全孔径相位误差函数的纽带,从而实现了基于相位梯度自聚焦(PGA)的精确的运动补偿。FFBP算法利用二维插值实现图像的递归融合,然而插值操作不可避免地产生插值误差,引起图像质量的损失。针对这个问题,第四章提出了一种基于波数谱融合的加速后向投影聚束SAR成像算法,即EBP算法。EBP算法创新地将子孔径数据后向投影到全局极坐标系,从而保证所有的子图像波数谱均位于同一波数空间。无需二维插值和递归融合,EBP算法仅需通过子图像波数谱的方位搬移便可得到全孔径波数谱。EBP算法精确地保留波数谱的原有形式,避免了二维插值处理带来的副作用,具有时域算法的精确性,而快速傅里叶变换(FFT)和循环移位操作又使其兼顾高效性。实验证明,EBP算法在图像质量和运算效率方面均优于FFBP算法。FFBP算法在聚束SAR领域取得了巨大成功,然而将其拓展至条带SAR处理还是一个很大的挑战。在第四章的启发下,第五章从波数谱的角度出发重新梳理了FFBP算法,找出了FFBP算法难以直接用于条带SAR处理的原因：第一、积分孔径；第二、角域升采样引起的大的运算负担。针对这个问题,第五章提出了重叠图像法,并成功实现了基于FFBP算法的条带SAR处理。该方法无需角域升采样,极大地保留了FFBP算法的运算效率优势,具有先聚束处理、后聚束一条带处理的特点。重叠图像法不但适用于正侧视或斜视条带SAR处理,还能拓展至TOPS SAR和滑动聚束SAR处理。本章最后,通过仿真实验和实测数据处理验证重叠图像的可行性和有效性。至此,线性孔径下不同成像模式皆可由快速时域算法实现。之前的章节实现了快速时域算法在线性孔径下的拓展应用,第六章针对圆迹SAR (CSAR)成像开展研究,提出了CEBP算法。为了避免过高的角域升采样率,CEBP算法将整个合成孔径(360°观测)划分为8个处理孔径。每个处理孔径单独开展孔径分解、子图像形成和波数谱融合,经由二维逆傅里叶变换(IFFT)得到处理孔径图像。将8幅处理孔径图像在直角坐标系下进行相干相加,得到最终聚焦的CSAR图像。CEBP算法继承了EBP算法精确和高效的优势,具备提供约λ/4的分辨能力。
[Abstract]:Synthetic Aperture Radar (SAR) imaging technology has the characteristics of all-weather, all-time, long-range, high-resolution, wide-area observation, which can effectively improve the radar information acquisition capability. It is widely used in civil areas such as land monitoring, ocean observation, sea ice monitoring and landform changes, as well as in the battlefield reconnaissance, military trend monitoring and other military areas. With the development of SAR technology, SAR is moving towards the direction of more flexible beam pointing, higher resolution and larger scene coverage, so it is important to obtain the surface information accurately. At the same time, each mode of operation can derive two geometric configurations: forward-sidelook and squint. Under different working modes and geometric configurations, the data acquisition mode and echo signal form are quite different, so it is necessary to study new imaging algorithms to meet the needs of different working modes. The main work of this paper is as follows: Chapter 2 is the basic theory of this paper, and introduces the current situation of the research. Several typical SAR imaging algorithms are discussed, and the imaging principle, key techniques, advantages and disadvantages of each algorithm are discussed. According to the characteristics of wavenumber support region, two resampling methods for strabismus spotlight SAR polar coordinate processing, namely LOSPI and SSPI, are introduced in the second chapter. It should be pointed out that PFA is not only tenacious, but also widely used in high resolution spotlight SAR, CSAR and video SAR. The imaging characteristics of range-Doppler focusing and LOSPI provide a reference for the self-focusing processing based on fast time-domain algorithm in Chapter 3. FBP algorithm and FFBP algorithm are used to discuss the dependence of image quality on the real APC position and terrain fluctuation. In practice, most motion errors can be corrected according to the platform motion information recorded by GPS/INS. Because of the influence of image focusing, it is necessary to develop a fast time-domain algorithm based on image or data self-focusing processing. In chapter 3, the FFBP algorithm is improved. Firstly, LOS virtual polar coordinate grid is selected to replace the polar coordinate grid in the original FFBP algorithm. Second, the overlapping sub-aperture configuration (OSF) is constructed by using the multi-aperture structure of FFBP algorithm. OSF is the link between the sub-aperture phase error and the full-aperture phase error function, thus realizing the phase-based phase error. The FFBP algorithm uses two-dimensional interpolation to realize the recursive fusion of images. However, interpolation inevitably produces interpolation errors, resulting in the loss of image quality. To solve this problem, the fourth chapter proposes an accelerated backward projection spotlight SAR imaging algorithm based on wavenumber spectrum fusion, namely EBP algorithm. The EBP algorithm innovatively projectes the sub-aperture data back to the global polar coordinate system to ensure that all the sub-image wavenumber spectra are in the same wavenumber space. Without two-dimensional interpolation and recursive fusion, the EBP algorithm can obtain the full-aperture wavenumber spectra only by the azimuth shift of the sub-image wavenumber spectra. Formal, avoiding the side effects of two-dimensional interpolation processing, has the accuracy of the time-domain algorithm, and fast Fourier transform (FFT) and cyclic shift operation make it both efficient. Experiments show that the EBP algorithm is superior to the FFBP algorithm in image quality and operational efficiency. FFBP algorithm has achieved great success in the field of spotlight SAR, but will be used in the future. Enlightened by Chapter 4, Chapter 5 reorganizes FFBP algorithm from the perspective of wavenumber spectrum and finds out the reasons why FFBP algorithm is difficult to be directly used in strip SAR processing: first, integral aperture; second, the heavy computational burden caused by angular domain rising sampling. In the fifth chapter, the overlapping image method is proposed and the stripe SAR processing based on FFBP algorithm is realized successfully. This method does not need angle-domain up-sampling and greatly retains the advantages of FFBP algorithm in operation efficiency. It has the characteristics of first spotlight processing and then spotlight-stripe processing. Finally, the feasibility and validity of overlapped images are verified by simulation experiments and real-time data processing. So far, different imaging modes in linear aperture can be implemented by fast time-domain algorithm. The previous chapters have realized the extended application of fast time-domain algorithm in linear aperture. Chapter 6 is aimed at the sixth chapter. In order to avoid high angle-domain sampling rate, the CEBP algorithm divides the whole synthetic aperture (360 degree observation) into eight processing apertures. Each processing aperture is decomposed separately, sub-image is formed and wavenumber spectrum is fused. The processing aperture map is obtained by two-dimensional inverse Fourier transform (IFFT). Image. Eight processed aperture images are added coherently in rectangular coordinates to obtain the final focused CSAR image. CEBP algorithm inherits the advantages of EBP algorithm, which is accurate and efficient, and has the ability to provide about a quarter of the resolution.
【学位授予单位】：西安电子科技大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TN957.52

【相似文献】