GPU Projection of ECAS-II Segmenter for Hyperspectral Images Based on Cellular Automata

Segmentation is a key issue in the processing of multidimensional images such as those in the field of remote sensing. Most of the segmentation algorithms developed for multidimensional images begin by reducing the dimensionality of the images, thus loosing information that could be relevant in the segmentation process. Evolutionary cellular automata segmentation (ECAS-II) is an evolutionary approach that provides cellular automata-based segmenters considering all the spectral information contained in a hyperspectral image without applying any technique for dimensionality reduction. This paper presents an efficient graphics processor unit implementation of the type of segmenters produced by ECAS-II for land cover hyperspectral images. The method is evaluated over remote sensing hyperspectral images, introducing it on a spectral–spatial classification scheme based on extreme learning machines. Experiments have shown that the proposed approach achieves better accuracy results for land cover purposes than other spectral–spatial classification techniques based on segmentation.

keywords: segmentation, Cellular automata (CA), CUDA, evolutionary cellular automata segmentation (ECAS-II), extreme learning machines (ELM), graphics processor unit (GPU), hyperspectral images