GPU用于高光谱数据高性能计算的应用实践与分析

许宁; 肖新耀; 胡玉新; 温静; 汪大明

GPU用于高光谱数据高性能计算的应用实践与分析

许宁^{1, 2, 3,},
肖新耀^{1, 2, 3},
胡玉新^{1, 2},
温静⁴,
汪大明⁵

1.
中国科学院空间信息处理与应用系统技术重点实验室, 北京 100190
2.
中国科学院电子学研究所, 北京 100190
3.
中国科学院大学, 北京 100049
4.
中国国土资源航空物探遥感中心, 北京 100083
5.
中国地质调查局油气资源调查中心, 北京 100029

基金项目:

中国地质调查局地质调查项目"地质勘查遥感系统集成与综合应用示范" 1212011120226

详细信息

作者简介:
许宁(1982-), 男, 四川邛崃人, 博士生, 研究方向光学遥感影像配准、融合处理及信息提取。E-mail:x_ning@aliyun.com

中图分类号: TP79
计量
- 文章访问数: 135
- HTML全文浏览量: 91
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2014-11-13
- 刊出日期: 2015-06-28

VALIDATION AND ANALYSIS OF HIGH PERFORMANCE COMPUTATION ON HYPERSPECTRAL IMAGERY BASED ON GPU

XU Ning^{1, 2, 3
,},
XIAO Xin-yao^{1, 2, 3},
HU Yu-xin^{1, 2},
WEN Jing⁴,
WANG Da-ming⁵

1.
Key Laboratory of Technology in Geo-spatial Information Processing and Application System, IECAS, Beijing 100190, China
2.
Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China
3.
University of Chinese Academy of Sciences, Beijing 100049, China
4.
China Aero Geophysical Survey and Remote Sensing Centre for Land and Resources, Beijing 100083, China
5.
Oil & Gas Survey, China Geological Survey, Beijing 100029, China

摘要

摘要: 高光谱遥感数据具有波段多、数据量大、处理复杂等特点, 基于GPU的高性能计算在遥感领域得到了快速发展, 为高光谱数据的快速处理提供了硬件和技术条件。采用GPU对高光谱遥感数据常用的SAM、PPI等处理算法进行应用实验, 验证基于GPU的高光谱遥感数据快速处理技术。实验采用新疆东天山地区的一景星载Hyperion数据, 利用支持IDL开发语言的GPULib、CUDA运行时API库进行算法效率的验证, 结果表明, 基于GPU的高光谱数据处理效率比常规的多核CPU主机处理效率有较大提升, 具有一定的应用推广价值。
- 高光谱数据 /
- GPU /
- 高性能计算 /
- SAM /
- PPI
Abstract: Hyperspectral imagery has many characteristics, such as plenty of bands, large volume of data, high computing complexity. In recent years, high performance computation has been making great progress in remote sensing based on GPU, providing the hardware and technical conditions for the rapid processing of hyperspectral data. We implemented the experiments on a hyperspectral image which was obtained by Hyperion of EO-1 satellite in East Tianshan area, Xinjiang, using SAM and PPI algorithms based on CPU and GPU, trying to study the fast processing technology on hyperspectral data.. Actually the GPULib and CUDA API were used through IDL language and the data was tested by different algorithms. The results show that the processing efficiency of hyperspectral data in GPU is greater than CPU and the technology can be used in remote sensing image processing.
- hyperspectral data /
- GPU /
- high performance computation /
- SAM /
- PPI

HTML全文

图 1 工作站主机和GPU组成的工作环境

Figure 1. Sketch of work environment constructed by Host and GPU

下载: 全尺寸图片幻灯片

图 2 基于单颗GPU实现的SAM处理流程图示

Figure 2. Flowchart of implementation for SAM algorithm on single GPU

下载: 全尺寸图片幻灯片

图 3 PPI算法中像元投影极值计数原理

Figure 3. Principle of PPI algorithm for projection of pixel vectors on skewers

下载: 全尺寸图片幻灯片

图 4 进行PPI实验的处理数据和处理结果

Figure 4. Hyperspectral data and results of PPI algorithm on CPU and GPU

下载: 全尺寸图片幻灯片

表 1 CPU-GPU基本性能测试

Table 1. Basic performance comparison of CPU and GPU

测试项目	数据大小	测试结果
测试项目	数据大小	矩阵乘法	矩阵转置	FFT计算
Q6600时间/s	5000×5000	126.859	0.000	4.047
Q6600时间/s	10000×10000	1626.453	0.001	23.121
C2050时间/s	5000×5000	0.157	0.016	0.125
C2050时间/s	10000×10000	0.452	0.032	0.405
处理时间对比	5000×5000	820.758	0.000	32.376
处理时间对比	10000×10000	3598.347	0.031	57.089

下载: 导出CSV

表 2 CPU与GPU的SAM性能测试

Table 2. Performance comparison of SAM algorithm between CPU and GPU

测试项目	5条参考光谱		200条参考光谱
测试项目	SAM计算时间	矩阵重列时间	SAM计算时间	矩阵重列时间
CPU/s	0.469	1.000	14.093	5.516
GPU/s	0.188	0.094	2.297	1.032
效率对比	2.500	10.638	6.135	5.345
平均效率	5.209		5.890

下载: 导出CSV

表 3 测试程序PPI与ENVI性能测试(CPU)

Table 3. Performance comparison of PPI between test program and ENVI on CPU

测试项目	1000单位随机向量		10000单位随机向量
测试项目	逐像元	分块优化	逐像元	分块优化
ENVI/s	245	13	2163	125
程序/s	501.641	183.200	5054.172	1320.618
效率对比	0.489	0.071	0.428	0.095

下载: 导出CSV

表 4 CPU与GPU测试程序PPI性能对比

Table 4. Performance comparison of PPI using test program between CPU and GPU

测试项目	1000单位随机向量		10000单位随机向量
测试项目	逐像元	分块优化	逐像元	分块优化
CPU/s	501.641	183.20	5054.172	1320.618
GPU/s	523.102	10.324	5128.320	95.682
效率对比	0.958974	17.745	0.985541	13.802

下载: 导出CSV

参考文献(15)

[1]	王润生, 甘甫平, 闫柏琨, 等.高光谱矿物填图技术与应用研究[J].国土资源遥感, 2010, (1):1~13. doi: 10.6046/gtzyyg.2010.01.01 WANG Run-sheng, GAN Fu-ping, YAN Bai-kun, et al. Hyperspectral mineral mapping and its application[J]. Remote Sensing for Land & Resources, 2010, (1): 1~13. doi: 10.6046/gtzyyg.2010.01.01
[2]	王润生, 熊盛青, 聂洪峰, 等.遥感地质勘查技术与应用研究[J].地质学报, 2011, 85(11):1699~1743. http://www.cnki.com.cn/Article/CJFDTOTAL-SDGJ201617057.htm WANG Run-sheng, XIONG Sheng-qing, NIE Hong-feng, et al. Remote sensing technology and its application in geological exploration[J]. Acta Geological Sinica, 2011, 85(11): 1699~1743. http://www.cnki.com.cn/Article/CJFDTOTAL-SDGJ201617057.htm
[3]	程宾洋. 高光谱遥感蚀变矿物填图算法并行设计与实现[D]. 成都: 成都理工大学, 2013. http://cdmd.cnki.com.cn/Article/CDMD-10616-1013263591.htm CHENG Bin-yang. The parallel design and implementation of hyperspectral remote sensing mineral mapping algorithm[D]. Chengdu: Chengdu University of Technology, 2013. http://cdmd.cnki.com.cn/Article/CDMD-10616-1013263591.htm
[4]	Gillis D, Bowles J H. Parallel implementation of the ORASIS algorithm for remote sensing data analysis[C]//Plaza A J, Chang C I. High performance computing in remote sensing. US: Taylor & Francis Group, 2008.
[5]	Tilton J C. Parallel implementation of the recursive approximation of an unsupervised hierarchical segmentation algorithm[C]// Plaza A J, Chang C I. High performance computing in remote sensing. US: Taylor & Francis Group, 2008.
[6]	Wang Jianwei, Chang Chein-I. FPGA design for real-time implementation of constrained energy minimization for hyperspectral target detection[C]// Plaza A J, Chang C I. High performance computing in remote sensing. US: Taylor & Francis Group, 2008.
[7]	Setoain J, Prieto M, Tenllado C, et al. Parallel morphological endmember extraction using commodity graphics hardware[J]. IEEE Geoscience and Remote Sensing Letters, 2007, 4(3): 441~445. doi: 10.1109/LGRS.2007.897398
[8]	Agathos A, Li J, Petcu D, et al. Multi-GPU implementation of the minimum volume simplex analysis algorithm for hyperspectral unmixing[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, 7(6): 2281~2296. doi: 10.1109/JSTARS.2014.2320896
[9]	杨靖宇, 张永生, 董广军.基于GPU的遥感影像SAM分类算法并行化研究[J].测绘科学, 2010, 35(3):9~11. http://www.cnki.com.cn/Article/CJFDTOTAL-CHKD201003003.htm YANG Jing-yu, ZHANG Yong-sheng, DONG Guang-jun. Investigation of parallel method of RS image SAM algorithmic based on GPU[J]. Science of Surveying and Mapping, 2010, 35(3): 9~11. http://www.cnki.com.cn/Article/CJFDTOTAL-CHKD201003003.htm
[10]	罗耀华, 郭科, 赵仕波.基于GPU的高光谱遥感MNF并行方法研究[J].四川师范大学学报:自然科学版, 2013, 36(3):476~479. http://www.cnki.com.cn/Article/CJFDTOTAL-SCSD201303036.htm LUO Yao-hua, GUO Ke, ZHAO Shi-bo. Minimum noise fraction of hyperspectral remote sensing in parallel computing based on GPU[J]. Journal of Sichuan Normal University: Natural Science, 2013, 36(3): 476~479. http://www.cnki.com.cn/Article/CJFDTOTAL-SCSD201303036.htm
[11]	宋义刚, 叶舜, 吴泽彬, 等.基于GPU的高光谱遥感图像PPI并行优化[J].航天返回与遥感, 2014, 35(4):74~80. http://www.cnki.com.cn/Article/CJFDTOTAL-HFYG201404011.htm SONG Yi-gang, YE Shun, WU Ze-bin, et al. Parallel optimization of Pixel Purity Index algorithm based on GPU for hyperspectral remote sensing image[J]. Spacecraft Recovery & Remote Sensing, 2014, 35(4): 74~80. http://www.cnki.com.cn/Article/CJFDTOTAL-HFYG201404011.htm
[12]	Kruse F A, Lefkoff A B, Boardman J W, et al. The Spectral Image Processing System (SIPS): Interactive visualization and analysis of imaging spectrometer data[J]. Remote Sensing of Environment, 1993, 44: 145~163. doi: 10.1016/0034-4257(93)90013-N
[13]	Boardman J W. Geometric mixture analysis of imaging spectrometery data[J]. Proc. Int. Geoscience and Remote Sensing Symp, 1994, 4: 2369~2371. http://ieeexplore.ieee.org/document/399740/
[14]	许宁, 胡玉新, 雷斌, 等.一种基于PPI的高光谱数据矿物信息自动提取方法[J].测绘科学, 2013, 38(4):138~141. http://www.cnki.com.cn/Article/CJFDTOTAL-CHKD201304046.htm XU Ning, HU Yu-xin, LEI Bin, et al. Automated mineral information extraction based on PPI algorithm for hyperspectral image[J]. Science of Surveying and Mapping, 2013, 38(4): 138~141. http://www.cnki.com.cn/Article/CJFDTOTAL-CHKD201304046.htm
[15]	Green A A, Berman M, Switzer P, et al. A transformation for ordering multispectral data in terms of image quality with implications for noise removal [J]. IEEE Transactions on Geoscience and Remote Sensing, 1988, 26(1): 65~74. doi: 10.1109/36.3001