Implementación en GPU del algoritmo K-Means para ... - UMBC

Implementación en GPU del algoritmo K-Means para procesamiento paralelo
de imágenes de satélite disponibles en la herramienta Google Maps
Abstract
In this work, we develop a new parallel implementation of the K-Means
unsupervised clustering algorithm for commodity graphic processing units (GPUs),
and further evaluate the performance of this newly developed algorithm in the task of
classifying (in unsupervised fashion) satellite imagery available from Google Maps
engine. Those images are obtained using a companion tool developed by the author
of this research work. With the ultimate goal of evaluating the classification
precision of the newly developed algorithm, we have analyzed the consensus or
agreement in the classification achieved by our implementation and an alternative
implementation of the algorithm available in commercial software (Research
Systems ENVI). Our experimental results, conducted using satellite images obtained
from Google Maps engine over different locations around the Earth, indicate that the
classification agreement between our parallel version and the ENVI implementation
of the K-Means algorithm is very high. In addition, the parallel version (developed
using the CUDA language available from NVidia) is much faster that the serial one
(more than 30x speedup), thus indicating that our proposed implementation can
significantly improve the computational performance of this clustering algorithm and
thus allows for larger scale processing of high-dimensional image databases such as
those available in the Google Maps engine used for validating the proposed GPU
implementation.
Keywords
K-Means, satellite imagery, GPU, CUDA, ENVI.
Trabajo Fin de Máster -9- Sergio Bernabé García