SAR快速成像与目标检测方法及GPU实现

发布时间：2018-04-23 04:24

本文选题：SAR成像 + GPU　；参考：《南京理工大学》2017年硕士论文

【摘要】：合成孔径雷达(SAR)不受时间、天气、地域等因素的影响,可以全天候工作,作用距离远,广泛应用于目标探测、识别和定位,战场侦察监视,自然灾害预报,资源勘探等军事和民用领域。SAR可以获得高分辨图像,但是成像算法复杂,需要处理大量的回波数据,运算量非常大,对硬件平台提出了很高的要求。本文针对星载SAR在轨成像处理的应用需求,开展了 SAR快速成像、目标检测与成像评估方法的研究,主要内容包括:(1)分析了星载SAR信号处理系统的设计因素,给出了一种以嵌入式GPU为核心的星载SAR信号处理系统,分析了 GPU并行处理效率,探讨了 CUDA平台的编程模型及存储模型。(2)详细讨论了 SAR回波模型,设计了 SAR回波信号存储方式;针对Chirp Scaling(CS)成像算法进行了并行处理的结构优化;针对NVIDIA GPU信号处理平台,设计了 CS成像算法并行程序;实验结果表明,相对于CPU平台,该系统取得了几十甚至几百倍的加速比。(3)针对双参数CFAR算法进行了相应的并行处理结构优化,设计了双参数CFAR算法的并行程序。利用真实的舰船SAR图像进行了验证,实验结果表明该方法目标检测准确,与CPU平台进行对比,加速比达到了 400以上。(4)详细讨论了 SAR图像质量客观评价指标,利用C#设计了 SAR图像质量客观评价的GUI测试软件,并对本文给出的星载SAR快速成像平台、传统CPU成像平台所成SAR图像质量进行了对比分析。
[Abstract]:The synthetic Aperture Radar (SAR), which is not affected by time, weather, region and other factors, can work around the clock and has a long range of functions. It is widely used in target detection, identification and positioning, battlefield reconnaissance and surveillance, and natural disaster prediction. High resolution images can be obtained from SAR in military and civil fields such as resource exploration, but the imaging algorithms are complex and need to deal with a large amount of echo data. In order to meet the requirements of spaceborne SAR imaging processing in orbit, this paper studies the methods of SAR fast imaging, target detection and imaging evaluation. The main content of this paper is to analyze the design factors of spaceborne SAR signal processing system. A space-borne SAR signal processing system based on embedded GPU is presented. The efficiency of GPU parallel processing is analyzed. The programming model and storage model of CUDA platform are discussed. The SAR echo model is discussed in detail. The storage mode of SAR echo signal is designed, the structure of parallel processing is optimized for Chirp scaling CSC imaging algorithm, the parallel program of CS imaging algorithm is designed for NVIDIA GPU signal processing platform, and the experimental results show that, compared with CPU platform, a parallel program for CS imaging algorithm is designed. The system achieves a speedup ratio of several tens or even hundreds of times. The parallel processing structure of the two-parameter CFAR algorithm is optimized and the parallel program of the two-parameter CFAR algorithm is designed. The real ship SAR image is used to verify the method. The experimental results show that the method is accurate and compared with CPU platform. The speedup ratio is more than 400. 4) the objective evaluation index of SAR image quality is discussed in detail. The GUI test software for objective evaluation of SAR image quality is designed by using C #, and the SAR image quality of Spaceborne SAR fast imaging platform and traditional CPU imaging platform is compared and analyzed in this paper.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【参考文献】