Parallelization of the K-means algorithm on a reconfigurable system: Application to hyper-spectral images

摘要

The article presents a parallel architecture dedicated to the k-means algorithm used for clustering objects in a nonudhierachical way. We propose an implementation on a reconfigurable system made of a PC and a FPGA board closelyudcoupled through the I/O bus. Experimentations have been carried on a set of hyper-spectral images and showudthat the computation time is reduced from a few hours to a few minutes. We also points out the influence of theudchannel quality between the processor and the FPGA board over the global system performances.