Implementación en GPU del algoritmo K-Means para ... - UMBC
Implementación en GPU del algoritmo K-Means para ... - UMBC
Implementación en GPU del algoritmo K-Means para ... - UMBC
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>Implem<strong>en</strong>tación</strong> <strong>en</strong> <strong>GPU</strong> <strong>del</strong> <strong>algoritmo</strong> K-<strong>Means</strong> <strong>para</strong> procesami<strong>en</strong>to <strong>para</strong>lelo<br />
de imág<strong>en</strong>es de satélite disponibles <strong>en</strong> la herrami<strong>en</strong>ta Google Maps<br />
Abstract<br />
In this work, we develop a new <strong>para</strong>llel implem<strong>en</strong>tation of the K-<strong>Means</strong><br />
unsupervised clustering algorithm for commodity graphic processing units (<strong>GPU</strong>s),<br />
and further evaluate the performance of this newly developed algorithm in the task of<br />
classifying (in unsupervised fashion) satellite imagery available from Google Maps<br />
<strong>en</strong>gine. Those images are obtained using a companion tool developed by the author<br />
of this research work. With the ultimate goal of evaluating the classification<br />
precision of the newly developed algorithm, we have analyzed the cons<strong>en</strong>sus or<br />
agreem<strong>en</strong>t in the classification achieved by our implem<strong>en</strong>tation and an alternative<br />
implem<strong>en</strong>tation of the algorithm available in commercial software (Research<br />
Systems ENVI). Our experim<strong>en</strong>tal results, conducted using satellite images obtained<br />
from Google Maps <strong>en</strong>gine over differ<strong>en</strong>t locations around the Earth, indicate that the<br />
classification agreem<strong>en</strong>t betwe<strong>en</strong> our <strong>para</strong>llel version and the ENVI implem<strong>en</strong>tation<br />
of the K-<strong>Means</strong> algorithm is very high. In addition, the <strong>para</strong>llel version (developed<br />
using the CUDA language available from NVidia) is much faster that the serial one<br />
(more than 30x speedup), thus indicating that our proposed implem<strong>en</strong>tation can<br />
significantly improve the computational performance of this clustering algorithm and<br />
thus allows for larger scale processing of high-dim<strong>en</strong>sional image databases such as<br />
those available in the Google Maps <strong>en</strong>gine used for validating the proposed <strong>GPU</strong><br />
implem<strong>en</strong>tation.<br />
Keywords<br />
K-<strong>Means</strong>, satellite imagery, <strong>GPU</strong>, CUDA, ENVI.<br />
Trabajo Fin de Máster -9- Sergio Bernabé García