文献类型：期刊文献

中文题名：海量超声数据体可视化研究

英文题名：Research of Large Ultrasonic Data Visualization

作者：潘卫国[1];何宁[2];薛健[1];吕科[1];翟锐[1];代双凤[1]

第一作者：潘卫国

通讯作者：Lü, Ke

机构：[1]中国科学院大学工程管理与信息技术学院;[2]北京联合大学信息学院

第一机构：中国科学院大学工程管理与信息技术学院,北京100049

年份：2016

卷号：44

期号：2

起止页码：472-478

中文期刊名：电子学报

外文期刊名：Acta Electronica Sinica

收录：CSTPCD;;EI(收录号：20161602257983);Scopus(收录号：2-s2.0-84963571291);北大核心:【北大核心2014】；CSCD:【CSCD2015_2016】；

基金：国家自然科学基金(No.U1301251;No.61271435);北京市自然科学基金(No.4141003)

语种：中文

中文关键词：体绘制;图形处理器;核外技术;并行计算;海量数据

外文关键词：volume rendering; graphics processing unit（GPU）; out-of-core; parallel computing; large data

摘要：近年来,随着科学数据的快速增长,海量数据的可视化分析成了急需解决的难题.越来越多的处理海量数据的方法向着并行、分布式处理的方向发展.本文提出了一种混合的框架来处理海量的超声数据,该框架通过整合多种硬件环境和计算资源来处理海量数据;所有的数据都存放在一个基于高速网络环境的数据共享中心,具有高性能显卡的前端工作站将耗时的处理任务分配到网络中的计算结点,而自身处理显示和交互的操作;同时基于Open CL和Open MP实现了可视化算法在GPU和CPU上的并行计算;核外算法应用在本框架中来处理海量的体数据.实验结果表明,本文提出的框架不仅可以处理海量数据,而且具有较高的交互性能.
In recent years,with the rapid growth of scientific data,large data analysis has become urgent problems.More and more large-data processing methods are modified to perform computation under parallel and distributed computing environment. In this paper,we present a hybrid architecture for large volume data visualization and processing. Various hardware environments and technologies are integrated in this architecture to perform interactive operations on very large volume datasets. All the datasets are stored in a data center with a gigabit network environment. The time-consuming data processing tasks are dispatched to the computing nodes connected to the same network,while the visualization and interaction operations are executed on a high-performance graphics workstation. Open CL and Open MP are used to implement volume rendering algorithms for accelerating visualization of a hierarchical volume data structure by both GPU and CPU with multi-cores,and some out-of-core algorithms are also presented to process the large dataset directly. The experimental results and practical application indicate that the hybrid architecture and methods presented in this paper are effective and efficient for the processing and visualization of very large volume datasets.