遥感影像融合的GPU加速及性能分析  

GPU-based Acceleration for Pansharpening Algorithms and Performance Analysis

作　　者：杨景辉[1] 张继贤[1] 

机构地区：[1]中国测绘科学研究院,北京100830

出　　处：《小型微型计算机系统》2016年第3期603-607,共5页Journal of Chinese Computer Systems

基　　金：国家"八六三"高技术研究发展计划项目(2011AA120401)资助;国家自然科学基金项目(40901229)资助

摘　　要：遥感影像融合是一种典型的数据密集型和计算密集型运算,常用的串行处理效率较低,研究GPU支持的并行处理方法有一定的现实意义.该文以NVIDIA公司新一代Fermi架构GPU为计算平台,采用CUDA并行计算环境,以SFIM融合算法为例,提出适用于遥感影像融合的并行处理方法.SPOT 5和Quick Bird两组大幅面遥感影像融合并行处理实验结果显示,提出的基于分块的核外计算方法,通用性强,易于实现,加速性能明显,根据整景影像测试结果,采用目前中等级别的GPU加速卡计算性能可提高至107倍.同时,该方法适用性广,可应用于多个影像融合算法.实验结果分析表明,除了GPU卡本身性能外,影响GPU加速性能主要因素包括单次导入影像的计算规模和主机与GPU之间交换数据量.Pansharpening algorithms are data- and computation-intensive,and their performance using serial computing is poor. Study of GPU-based acceleration is a realistic concern. In this paper,we present a parallel processing method based on GPU for pansharpening algorithms,taking SFIM（ Smoothing Filter Based Intensity Modulation） fusion algorithm as an example. The method exploits CUDA（ Compute Unified Device Architecture） and a GPU card with NVIDIA＇ s Fermi architecture. The results of two experiments on SPOT 5 and Quick Bird images,respectively,both with large frames showthat the methodr unning on a GPU card currently being middle-performance can achieve speedups up to 107 ×. The proposed out-of-core method based on block partition has broad applicability,itis easy to implement and is obvious computational acceleration. Moreover,the method can be adapted to several pansharpening algorithms. Through analyzing the experimental results,we find that the factors impacting the speedup are the computational size of thedata block imported into GPU card,the amount of exchanged data between host computer and GPU card,and the performance of the used GPU card.

关 键 词：遥感影像 融合 图形处理器 并行计算 核外计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]