Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/10197
標題: 利用圖形處理器加速高光譜影像主成份分析之研究
GPU-Accelerated Principal Component Analysis for Hyperspectral Images
作者: 蔡家瑋
Tsai, Chia-Wei
關鍵字: 主成分分析;PCA;圖形處理器;CUDA;平行計算;高光譜影像;GPU;CUDA;Parallel Computing;Hyperspectral Image
出版社: 土木工程學系所
引用: 徐百輝,2007。大地的辨識密碼-高光譜影像,科學發展,第416期,第13-19頁。 張舒、褚豔利、趙開勇、張鈺勃,2009。GPU高性能運算之CUDA,中國水利水電出版社。 蔡榮得,2011。影像處理授課教材,國立中興大學土木工程學系。 林俊淵、周嘉奕、林郁翔、李昇達、陳昱蓉、黃宣穎、李天齡,2011,CUDA 輕鬆上手─新世代 GPU 應用技術,松崗出版社。 羅竣威、謝清祿,2005。三種主軸成分法於數位影像特徵擷取與判別之研究,第四屆數位典藏技術研討會,中央研究院資訊科學研究所,2005年9月1-2日。 Burden, R. L. and Faires, J. D., 2011. Numerical Analysis, 9th ed., Brooks/Cole Cengage Learning. Du, P., Parsons, M., Fuentes, E., Shaw, S. L. and Dongarra, J., 2010. Tuning Principal Component Analysis for GRASS GIS on Multi-core and GPU Architectures, FOSS4G 2010, Barcelona, Spain, Sept. 6-9, 2010. EARDAS, 1991. EARDAS Field Guide. Francis, J. G. F., 1961. “The QR Transformation,” The Computer Journal, vol. 4, no. 3, pp. 265-271. Gonzalez, R. C. and Woods, R. E., 2008. Digital Image Processing, 3rd ed., Prentice Hall. Green, R. O., Eastwood, M. L., Sarture, C. M., Chrien, T. G., Aronsson, M., Chippendale, B. J., Faust, J. A., Pavri, B. E., Chovit, C. J., Solis, M. and Olah, M. R., 1998. “Imaging Spectroscopy and the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS)¸” Remote Sensing Environment, Vol. 65, pp. 227-248. Harris, M., 2007. Parallel prefix sum (Scan) with CUDA, NVIDIA. Hoare, C. A. R., 1962. “Quicksort,” The Computer Journal, vol. 5, no. 1, pp. 10-16.. Hotelling, H., 1933. “Analysis of a complex of statistical variables into principal components,” Journal of Educational Psychology, vol. 24, no. 6, pp. 417-441. Intel, 2012. http://ark.intel.com/zh-tw/compare/63696, accessed March 13, 2012. NVIDIA, 2011a. NVIDIA CUDA C Programming Guide Version 4.1. NVIDIA, 2011b. NVIDIA GPU computing SDK Version 4.1. Pearson, K., 1901. “On Lines and Planes of Closest Fit to Systems of Points in Space,” Philosophical Magazine, vol. 2, pp. 559-572. Quintiliano, P. and Santa-Rosa, A., 2003. “Detection of streets based on KLT using IKONOS multispectral images,” 2nd GRSS/ISPRS Joint Workshop on Remote Sensing and Data Fusion over Urban Areas, Berlin, Germany, pp. 186-190. Schowengerdt, R. A., 1997. Remote Sensing: Models and Methods for Image Processing, Academic Press. Sirovich, L. and Kirby, M., 1987. “Low-dimensional procedure for the characterization of human faces,” Journal of Optical Society of America A, vol. 4, no. 3, pp. 519-524. Turk, M. A. and Pentland, A. P., 1991. “Face recognition using eigenfaces,” Proc. CVPR ''91 (Computer Vision and Pattern Recognition, 1991), Maui, Hawaii, June 03-06, pp. 586-591. Zhang, J. and Lim, K. H., 2011. “Implementation of a covariance-based principal component analysis algorithm with a CUDA-enabled graphics processing unit,” IGARSS 2011, Vancouvor, Canada, pp. 1759-1762
摘要: 
使用高光譜影像時,常會遇到鄰近波段影像間存在高相關性的問題,這些相關的波段傳遞了近似的資訊,此時常會使用主成份分析(Principal Component Analysis; PCA)將高相關性之資料轉換至獨立不相關的主軸空間,以減少後續分析應用波段維度之需求。鑑於圖形處理技術之發展,圖形處理器(Graphics Processing Unit; GPU)的運算速度及記憶體頻寬已遠超過中央處理器(Central Processing Unit; CPU),若以GPU來運算,必能有效提昇高光譜影像PCA的運算效率。
本研究旨在探討如何利用NVIDIA公司所提出的統一計算設備架構(Compute Unified Device Architecture; CUDA),設計以GPU運算的PCA程式(稱為GPU-PCA),並在相同的演算法之下,與使用CPU運算的單執行緒PCA程式(稱為CPU-PCA)比較運算效能。研究中採用以共變異數為基礎的PCA演算法,使用一張7個波段256×256像素的Landsat TM 衛星影像及一張191個波段1280×307像素的HYDICE(Hyperspectral Digital Imagery Collection Experiment)航空影像進行測試。實驗結果顯示,相較於CPU-PCA的運算速率,GPU-PCA對於數據量較小的Landsat TM衛星影像未達到加速效果,但對於數據量龐大的HYDICE航空影像,GPU-PCA有效利用GPU平行計算能力,整體加速比達30倍,效果極佳。

There is always encountered with problems of high correlation among neighboring spectral bands, which deliver similar spectral information, when using hyperspectral images. Techniques in principal component analysis (PCA) are usually used to transform these high correlation image data into independent space in order to reduce the demand of the number of bands in following computations. With the advanced development in graphics computing technology, the computing speed and memory bandwidth of the Graphics Processing Unit (GPU) are much faster than those of the Central Processing Unit (CPU). Thus, we can take the advantage of GPU computing to improve the computational efficiency of the PCA operation.
The purpose of this study is to apply NVIDIA’s Compute Unified Device Architecture (CUDA) in designing programs for PCA operations with the aid of GPU (named GPU-PCA), and to compare the computational efficiency with a single-thread PCA counterpart program which operates with CPU (named CPU-PCA). Both GPU-PCA and CPU-PCA programs were implemented with the covariance-based PCA algorithm. A 7-band 256×256 Landsat TM satellite image and a 191-band 1280×307 HYDICE (Hyperspectral Digital Imagery Collection Experiment) aerial image were used in the experiments. Experimental results show that the GPU-PCA accelerates PCA operation the more the large amount of the image data, i.e., less in the small-amount Landsat TM satellite image and more in the huge-amount HYDICE aerial image. In the HYDICE aerial image, it is shown that the GPU-PCA accelerates PCA operation in an overall speedup of 30 through the parallel computing capability of GPU.
URI: http://hdl.handle.net/11455/10197
其他識別: U0005-2308201206422600
Appears in Collections:土木工程學系所

Files in This Item:
File Description SizeFormat Existing users please Login
nchu-101-7099062075-1.pdf12.45 MBAdobe PDFThis file is only available in the university internal network    Request a copy
Show full item record
 
TAIR Related Article

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.