面向超大型栅格数据的并行快速傅里叶变换算法  

A Novel Extremely Large Raster Data Parallelized Fast Fourier Transform Algorithm

作　　者：潘欣[1,2] 杨典华[3] 张树清[2] 张策[2] 姜春雷[2] 崔树林[2] 

机构地区：[1]长春工程学院计算机技术与工程学院,吉林长春130012 [2]中国科学院东北地理与农业生态研究所,吉林长春130102 [3]首都师范大学三维信息获取与应用教育部重点实验室,北京100048

出　　处：《地理与地理信息科学》2013年第4期52-55,98,共5页Geography and Geo-Information Science

基　　金：国家863计划项目子课题(2011AA120302);国家自然科学基金青年基金项目(41101384)

摘　　要：傅里叶变换是地理栅格数据处理的重要算法之一。但当面对超大型栅格数据时,传统的单个进程方法无法快速地处理出结果;而且由于多数方法需一次加载所有数据,所以很多程序难以处理超过系统内存大小的数据。该文提出一种面向超大型栅格数据的并行快速傅里叶变换算法,通过"列暂存文件"作为交换空间,实现整个数据的分步骤处理,在有限的内存空间下,可以并行地分步骤、分块地处理整个数据。实验表明,该算法可以显著提高傅里叶变换运行速度,并一直占用较低的内存空间。Fourier transform is an important algorithm in the field of geographic data processing. However, when process extremely large raster data, the traditional single-process algorithm cannot product results quickly; and most method can＇t compatible with the data size more than the system memory size because they need load all the data in to memory. This paper presents a novel extremely large raster data parallelized fast Fourier transform algorithm,through ＂temporary column files＂ as exchange space, it can parallelize and partitioned process the entire raster data step-by-step with limited memory space. Experiments show, the proposed method can significantly improve the Fourier transform running speed,and only need relatively limited memory space when the program executing.

关 键 词：栅格 并行 傅里叶分析 超大型数据 MPI 

分 类 号：P208[天文地球—地图制图学与地理信息工程]