【作者】 周松涛；

【导师】 边馥苓；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2013， 博士

【摘要】 发展高分辨率对地观测技术是我国2006年确立的16个重大科技专项之一,这一项目的实施,必将促进我国高分辨率遥感应用技术的发展,同时也为遥感数据的应用与处理带来了挑战。在有条件能够获取高分辨率遥感数据的同时,必将需要一套更加高效的处理方法,并最终使之得到有效的应用,否则这一技术的发展就失去了根本。基于分布式的存储,通过高性能的集群对海量遥感数据进行处理,已经是一种非常有效的方式。但对于集群应用环境来说,成本,能耗,以及规模等问题,都可能成为制约处理效率方面的因素,近几年来,基于GPU的通用计算技术为我们在这方面提供了另外一种途径。作为专门为图形,图像处理提供硬件加速的硬件,其通用运算接口能够为影像处理提供加速几乎是自然而然的事情。采用GPU+CPU的混合计算系统,比以往单纯CPU运算高出几十倍甚至几百倍,上千倍,将一直局限在大型服务器集群和超型计算机领域的高性能计算推向主流,可以使得PC和工作站具有超级计算的能力。此外,空间信息的数字化存储,网络化传输,可视化表达和智能化应用是一个必然的趋势,而这几个方面是相互关联,存在制约关系的。好的数据存储方式,优化的网络传输,必然能够提高海量遥感数据的可视化应用效率。传统的存储方式往往存在弊端,有的模式是为数据管理和生产而设计,有的则是为可视化应用而设计。将二者有机统一起来,并采用一些新技术,在可视化技术中适当创新,也是本文讨论的问题之一。本文从数据的存储,处理及应用三个方面阐述了海量遥感数据在计算机中的应用问题,文中着重论述了如下几个部分：1)海量遥感数据的存储模式及调度结合遥感数据的生产处理系统与可视化应用系统对数据存储的需求特点,提出了一个混合模式的数据管理系统,在实际应用中取得了满意的效果,既满足了数据管理上的需求,又符合可视化应用系统的需要。2)遥感数据的高性能处理及分析这部分属于本文的重点内容。结合实际遥感数据处理算法的特点,论述了采用GPU加速技术,对算法进行改造优化的步骤,从而提高了算法效率。着重以基于相关系数的匹配算法,以及大范围条件下矢量与栅格之间的空间分析算法为代表,分别阐述了遥感数抓的邻域处理,矢量运算,矢量栅格联合分析以及全局数据统计等算法的处理流程和优化方式。3)海量遥感数据的三维可视化应用主要针对海量遥感数据的地形渲染及交互方式进行了叙述。重点介绍了数据库中任意索引格网划分的条件下,DEM和正射影像之间如何正确匹配渲染,仍然使用了基于可编程的GPU渲染技术,无论从效率,效果上,都优于传统的直接基于OpenGL的渲染方法。尤其是在可视化表达手段上,由于采用高度可编程的方式,复杂的可视化效果只需要简单的手段即可实现,对复杂多维地形相关信息的可视化表达提供了一个非常好的于段.上述几个组成部分是针对一个基于海量遥感数抓的可视化管理及应用系统,对其中影像效率的环节加以重点关注,并提出解决方法。文中叙述的方法已经在某集群数据处理系统中得到应用,而且取得了非常好的效果。尤其是通过GPU通用计算技术实现的栅格处理算法,较用户的原始算法都有几倍,几十倍的效率提升,大大提高了整个集群系统的处理效率。更多还原

【Abstract】 Development of high-resolution earth observation technology was16major projects for science and technology development established in2006in China, and the implementation of this project will promote the development of its application technology. A very effective way to fulfill this project is using some techniques such as distributed storage, high-performance cluster etc, but these techniques bring issues such as cost, energy consumption. Recent years, GPU-based general-purpose computing technology bring us another way to resolve the problem of large scale computing. In conditions of moderm GPU technology, using general-purpose GPU computing interface can provide acceleration for image processing efficiency, and it can provid at least several times higher performance than mormal computer systems by using hybrid GPU+CPU computing systems, even hundreds of times, thousands of times more.In addition, spatial information’s digital storage, transmission through network, visualizations and intelligent application is an inevitable trend, it will bring us new challenges. For good use these massive remote sensing data, there are many problems to be solved such as network transmission optimization, storage methods, application model etc, and these also are the key issues discussed in this article.This paper’s main content as follows:1) Storage model and scheduling of massive remote sensing dataTo meet the requirements of data processing system and visualization applications of remote sensing data, proposed a mixed-mode data management system, and achieved satisfactory results in practical applications..2) High-performance processing and analysis of remote sensing dataThis part is the focus of this article content. Combined with the actual characteristics of the remote sensing data processing algorithms, this part discusses the technique that using GPU acceleration technology to optimize image processing algorithm. Emphasis on surface correlation algorithm, as well as the analysis algorithm for vector and raster data in a wide range conditions, explained the convolution, image mathching, vector arithmetic, vector and raster joint analysis and global statistics process in detail.3) Massive remote sensing data visualization applicationIn this section, The author highlights the technology of3D terrain rendering and its interactive method for massive remote sensing data. Focuses on the conditions of vary index in the database of the DEM and the orthophoto, how to render them correct without coordinate registration. Also used the technique of programmable GPU, the experiment achieved satisfactory results in both of efficiency and effectiveness. Specifically in the complex visual effects applications, it is very simple than the traditional opengl way in implementation.The technique discussed in this paper is very important to the solutions for the management and application system based on massive remote sensing data visualization, and has been successful applied in a data processing system in cluster environment. Using the general-purpose GPU technology, some raster processing algorithms had achieved ten or several dozen times efficiency increased compared to the user’s original way.更多还原