TY - GEN
T1 - Recursive Orthogonal Vector Projection for Hyperspectral Image Abundance Estimation Based on GUP
AU - Yu, Chunyan
AU - Huang, Jin
AU - Song, Meiping
AU - An, Dong
AU - Chang, Chein I.
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2018/9
Y1 - 2018/9
N2 - Hyperspectral remote sensing data contain more material information for each endmember due to the complexity of the natural object and the limitation of spatial resolution, resulting in the existence of a large number of mixed pixels, which increases the difficulty of data analysis. Abundance estimation is one of the most important topics in hyperspectral unmixing, it can be used to analyze the proportion of mixed pixels accurately. In order to improve the processing speed of hyperspectral image abundance estimation, in this paper, the parallel mode of Recursive Orthogonal Vector Projection (ROVP) algorithm based on NVIDIA's graphic processing unit (GPU) is proposed. The ROVP-C (ROVP-on-CUDA) algorithm based on CPU / GPU heterogeneous system and the ROVP-L (ROVP-on- Library) algorithm based on CUBLAS (CUDA Basic Linear Algebra Subprograms) library are designed and implemented. The experimental results showed that these two algorithms have achieved obvious speed-up ratio compared with the traditional serial algorithms, and it showed that GPU has a great advantage in the field of estimating the hyperspectral abundance.
AB - Hyperspectral remote sensing data contain more material information for each endmember due to the complexity of the natural object and the limitation of spatial resolution, resulting in the existence of a large number of mixed pixels, which increases the difficulty of data analysis. Abundance estimation is one of the most important topics in hyperspectral unmixing, it can be used to analyze the proportion of mixed pixels accurately. In order to improve the processing speed of hyperspectral image abundance estimation, in this paper, the parallel mode of Recursive Orthogonal Vector Projection (ROVP) algorithm based on NVIDIA's graphic processing unit (GPU) is proposed. The ROVP-C (ROVP-on-CUDA) algorithm based on CPU / GPU heterogeneous system and the ROVP-L (ROVP-on- Library) algorithm based on CUBLAS (CUDA Basic Linear Algebra Subprograms) library are designed and implemented. The experimental results showed that these two algorithms have achieved obvious speed-up ratio compared with the traditional serial algorithms, and it showed that GPU has a great advantage in the field of estimating the hyperspectral abundance.
UR - http://www.scopus.com/inward/record.url?scp=85073907833&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85073907833&partnerID=8YFLogxK
U2 - 10.1109/WHISPERS.2018.8747232
DO - 10.1109/WHISPERS.2018.8747232
M3 - Conference contribution
AN - SCOPUS:85073907833
T3 - Workshop on Hyperspectral Image and Signal Processing, Evolution in Remote Sensing
BT - 2018 9th Workshop on Hyperspectral Image and Signal Processing
PB - IEEE Computer Society
T2 - 9th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing, WHISPERS 2018
Y2 - 23 September 2018 through 26 September 2018
ER -