| GPU上高光谱快速ICA降维并行算法 |
| |
| 引用本文： | 方民权,周海芳,张卫民,申小龙.GPU上高光谱快速ICA降维并行算法[J].国防科技大学学报,2015,37(4):65-70. |
| |
| 作者姓名： | 方民权 周海芳 张卫民 申小龙 |
| |
| 作者单位： | 国防科学技术大学 计算机学院,国防科学技术大学 计算机学院,国防科学技术大学 计算机学院,国防科学技术大学 计算机学院 |
| |
| 基金项目： | 国家自然科学基金项目（面上项目，重点项目，重大项目） |
| |
| 摘 要： | 高光谱影像降维快速独立成分分析过程包含大规模矩阵运算和大量迭代计算。通过分析算法热点,设计协方差矩阵计算、白化处理、ICA迭代和IC变换等关键热点的图像处理单元映射方案,提出并实现一种G-Fast ICA并行算法,并基于GPU架构研究算法优化策略。实验结果显示:在处理高光谱影像降维时,CPU/GPU异构系统能获得比CPU更高效的性能,G-Fast ICA算法比串行最高可获得72倍加速比,比16核CPU并行处理快4~6.5倍。
|
| 关 键 词： | 图像处理单元 高光谱影像降维 快速独立成分分析 并行算法 性能优化 |
| 收稿时间： | 2014/9/28 0:00:00 |
| 修稿时间： | 4/3/2015 12:00:00 AM |
A parallel algorithm of FastICA dimensionality reduction for hyperspectral image on GPU |
| |
| FANG Minquan,ZHOU Haifang,ZHANG Weimin and SHEN Xiaolong.A parallel algorithm of FastICA dimensionality reduction for hyperspectral image on GPU[J].Journal of National University of Defense Technology,2015,37(4):65-70. |
| |
| Authors: | FANG Minquan ZHOU Haifang ZHANG Weimin and SHEN Xiaolong |
| |
| Institution: | College of Computer, National University of Defense Technology, Changsha 410073, China,College of Computer, National University of Defense Technology, Changsha 410073, China,College of Computer, National University of Defense Technology, Changsha 410073, China and College of Computer, National University of Defense Technology, Changsha 410073, China |
| |
| Abstract: | Fast Independent Component Analysis (FastICA) dimensionality reduction for hyperspectral image needs a large amount of matrix and iterative computation. By analyzing the FastICA algorithm, we know that its hotspots are covariance matrix calculation, white processing, ICA iteration, and IC transformation. In the paper, we propose a GPU-oriented mapping scheme (G-FastICA), and then optimize the algorithm on memory accessing and computation-communication overlapping. We also investigate the performance impact of thread-block size. Our experimental results have shown that we obtain much better performance: the G-FastICA algorithm can reach a speedup of up to 72x compared to the sequential code on CPU, and it runs 4x~6.5x faster than the case when using a 16-core CPU. |
| |
| Keywords: | graphic processing units hyperspectral image dimensionality reduction fast independent component analysis parallel algorithm performance optimization |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《国防科技大学学报》浏览原始摘要信息 |
| 点击此处可从《国防科技大学学报》下载免费的PDF全文 |
|
