论文部分内容阅读
目的几何校正(又称地理编码)是合成孔径雷达(SAR)影像处理流程中重要的一个步骤,具有一定的计算复杂度,需要用到几何定位模型。本文针对星载SAR影像,采用有理多项式系数(RPC)定位模型,提出了图形处理器(GPU)支持的几何校正大规模并行处理方法。方法该方法充分利用GPU计算资源强大及几何校正过程中每个像素处理步骤一致的特点,每次导入大量像素至GPU,为每个像素分配一个线程,每个线程执行有理函数计算、投影变换、插值采样等计算复杂度高的步骤,通过优化配置dim Grid和dim Block参数,提升GPU的并行性能。该方法通过分块处理实现SAR影像大幅面处理,且可适用于多个不同分块大小。结果实验结果显示其计算加速比为38 44,为全面客观地分析GPU并行处理的特点,还计算了整体加速比,通过多个实验分析影响整体加速性能的因素,提出大块读写提高I/O性能的优化方法。结论该方法形式简洁,通用性好,可适用于几乎所有的星载SAR影像、不同的影像幅面;且加速性能明显。
The purpose of geometric correction (also known as geocoding) is an important step in synthetic aperture radar (SAR) image processing flow, with a certain degree of computational complexity, the need to use the geometric positioning model. In this paper, the positioning model of rational polynomial coefficients (RPC) is proposed for spaceborne SAR images, and a large scale parallel processing method of geometry correction supported by GPU is proposed. Method This method makes full use of the advantages of GPU computing resources and uniform processing steps of each pixel in the geometric correction process. Each time a large number of pixels are imported into the GPU, one pixel is allocated to each thread, and each thread performs calculation of rational functions, projection transformation, Interpolation, sampling and other computational complexity of the steps, through the optimal allocation of dim Grid and dim Block parameters to enhance the GPU’s parallel performance. The method achieves SAR image large format processing by block processing, and can be applied to a plurality of different block sizes. The experimental results show that the calculated speedup ratio is 38.44. In order to comprehensively and objectively analyze the characteristics of GPU parallel processing, the overall speedup ratio is also calculated, and the factors that affect the overall acceleration performance are analyzed through multiple experiments. O performance optimization methods. Conclusion The proposed method is simple in form and good in generality. It can be applied to almost all the spaceborne SAR images and different image formats; and the acceleration performance is obvious.