地球信息科学学报 ›› 2013, Vol. 15 ›› Issue (1): 128-136.doi: 10.3724/SP.J.1047.2013.00128

• 遥感技术与应用 • 上一篇    下一篇

海量遥感数据的高性能地学计算应用与发展分析

杨海平1,2, 沈占锋1, 骆剑承1, 吴炜1,2   

  1. 1. 中国科学院遥感应用研究所, 北京 100101;
    2. 中国科学院大学, 北京 100049
  • 收稿日期:2012-09-24 修回日期:2012-12-10 出版日期:2013-02-25 发布日期:2013-02-25
  • 作者简介:杨海平(1987-),女,硕士研究生,主要研究方向为高性能遥感地学计算与信息提取。E-mail:yanghp@irsa.ac.cn
  • 基金资助:

    国家自然科学基金项目(40971228,41101398);国家科技支撑计划项目(2011BAH06B02)。

Recent Developments in High Performance GeoComputation for Massive Remote Sensing Data

YANG Haiping1,2, SEHN Zhanfeng1, LUO Jiancheng1, WU Wei1,2   

  1. 1. Institute of Remote Sensing Applications, CAS, Beijing 100101, China;
    2. University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2012-09-24 Revised:2012-12-10 Online:2013-02-25 Published:2013-02-25

摘要:

航空及航天遥感器的快速发展,使得多源、多时空分辨率的遥感数据成TB级增长,对海量遥感数据的高性能计算与处理提出了更高的要求。据此,当前的遥感应用已经吸收了新型硬件架构计算、集群计算和分布式计算等高性能计算领域的最新技术。本文针对高性能计算处理海量遥感数据的效率问题,分别从分布式并行遥感文件系统和高性能遥感地学计算模式两个方面来论述该问题的研究进展;在此基础上,列举了当前具有代表性的集群和分布式遥感计算平台/系统,并结合具体实验工作,详细阐述了遥感高性能计算平台gDos-IPM(Geospatial Data Operation System-Image Processing Machine)的设计思路;最后总结了高性能遥感地学计算的发展趋势。

关键词: 分布式计算, 并行文件系统, 多核, 集群, 图形处理器

Abstract:

As the amount of remote sensing data is sharply increasing within the continuing development in remote sensors, the exploitation of massive amount of remote sensing data is booming in recent years. Therefore, the computational problems in the applications that involve the large collection of remotely sensed imagery processing, such as global climate change and hazard assessment, arise inevitably. On this point, high performance computing (HPC)-based patterns, including cluster computing, grid computing, cloud computing and computing with hardware such as field-programmable gate arrays (FPGA) and graphic processing units (GPU), are introduced to the applications that concern a huge amount of remote sensing data processing. This paper focuses on the state of the art coping with the challenges that emerge when the massive remote sensing data are processed by the HPC-based platforms. In particular, we review recent developments in the parallel file systems for storing the remote sensing data and high performance geocomputation. Specifically, the HPC-based paradigms delivered in this paper involve cluster-based platform, grid and cloud based environments. Further, the typical examples of the HPC-based platforms that process the massive remote sensing data, comprising the Pixel Factory, the Grid Processing on Demand (G-POD) and the Geospatial Data Operation System-Image Processing Machine (gDos-IPM), are discussed. And the gDos-IPM, a solution for the platform of high performance computing for remote sensing, is described in detail. The gDos-IPM, which integrates the computation and storage resources and involves GPUs, multicore processors and clusters, provides the remote sensing tools about preprocessing and information extraction for massive remote sensing data in the heterogeneous computing environment. Also, it supports a dynamic model for spatio-temporal data and multi-level parallel computing. At the end of this paper, we present a thoughtful view on the challenges of HPC for further remote sensing applications.

Key words: multicore, parallel file system, graphic processing unit (GPU), distributed computing, cluster