高分辨率机载SAR成像及GPU实现

发布时间：2018-06-15 02:52

本文选题：合成孔径雷达 + 图形处理器　；参考：《南京航空航天大学》2017年硕士论文

【摘要】：机载合成孔径雷达(Synthetic Aperture Radar,简称SAR)能够全天时、全天候地工作,并且具有高分辨率、广视域等众多优点,在多个领域得到广泛的应用。波数域算法是一种高精度的SAR处理算法。补偿回波的相位误差是保证成像质量的必要手段,因为相位误差具有二维空变性,因此需要对传统的自聚焦技术进行改进,以抑制相位误差的二维空变性。高速发展的图形处理器(Graphics Processing Unit,简称GPU)和统一计算架构(Compute Unified Device Architecture,简称CUDA)技术的发展,使得GPU的通用并行计算得到普及,成为了SAR成像算法的一种新型运算平台。基于以上所述,本文主要工作包括:(1)分析SAR成像模型,波数域成像算法原理,以及基于惯导测量数据的运动补偿原理,并分析算法的并行性,设计该算法基于GPU平台的并行方案。点目标仿真数据处理结果显示该算法在GPU平台上可获得30倍左右的加速比,验证了并行方案的可行性。(2)针对条带SAR相位误差的方位向空变性,讨论了一种子孔径自聚焦算法——PGA-MD算法。该算法通过PGA算法计算子孔径的精确的相位误差估计,并利用MD算法校正误差估计中的线性相位,最后将子孔径相位误差拼接得到全孔径相位误差。并且分析PGA-MD算法的并行性,设计并行方案。最后通过实测数据处理,验证了PGA-MD算法的有效性以及并行方案的可行性。(3)针对相位误差的距离向空变性,提出改进的距离向分块自聚焦处理技术。该算法在距离向进行部分重叠的分块,并对每个距离块进行PGA-MD相位误差估计和补偿以及子图像自聚焦处理,得到每个距离块的成像结果。并且本文提出了子图像互相关配准算法来实现相邻距离块的配准,并用加权拼接算法对配准后的图像进行拼接,得到完整的成像结果。实测数据处理结果表明,距离向分块自聚焦算法能够进一步提高聚焦效果。
[Abstract]:Airborne synthetic Aperture Radar (SAR) can work all day, all weather, and has many advantages such as high resolution, wide view and so on, so it has been widely used in many fields. Wavenumber domain algorithm is a high precision SAR processing algorithm. Compensating the phase error of echo is a necessary means to guarantee the imaging quality, because the phase error has two-dimensional space variability, so it is necessary to improve the traditional self-focusing technique to suppress the two-dimensional null variation of phase error. With the development of Graphics processing Unit (GPU) and Unified Compute Unified device Architecture (CUDAE), GPU has become a new computing platform for SAR imaging. Based on the above, the main work of this paper is to analyze SAR imaging model, wavenumber domain imaging algorithm and motion compensation principle based on inertial navigation measurement data, and analyze the parallelism of the algorithm. The parallel scheme based on GPU platform is designed. The result of point target simulation data processing shows that the algorithm can get about 30 times speedup on GPU platform, which verifies the feasibility of the parallel scheme. A subaperture autofocus algorithm, PGA-MD algorithm, is discussed. The PGA algorithm is used to calculate the accurate phase error estimation of the sub-aperture and the MD algorithm is used to correct the linear phase in the error estimation. Finally, the sub-aperture phase error is spliced to obtain the full aperture phase error. The parallelism of PGA-MD algorithm is analyzed and the parallel scheme is designed. Finally, the validity of the PGA-MD algorithm and the feasibility of the parallel scheme are verified by the experimental data processing. (3) in view of the range spatial variability of the phase error, an improved range block autofocus processing technique is proposed. The algorithm partially overlaps blocks in the range direction and performs PGA-MD phase error estimation and compensation for each distance block as well as sub-image self-focusing processing to obtain the imaging results of each range block. In this paper, a sub-image cross-correlation registration algorithm is proposed to realize the registration of adjacent distance blocks, and the image is stitched by weighted stitching algorithm, and the complete imaging results are obtained. The experimental data processing results show that the range block autofocus algorithm can further improve the focusing effect.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【参考文献】