Creating and sharing knowledge for telecommunications

SISAL Implementation on GPU for Hyperspectral Unmixing

Cedillo, J. ; Bioucas-Dias, J. ; Silva, V. ; Nascimento, J.

SISAL Implementation on GPU for Hyperspectral Unmixing, Proc Conf. on Telecommunications - ConfTele, Aveiro, Portugal, Vol. 0, pp. 0 - 0, September, 2015.

Digital Object Identifier:

Download Full text PDF ( 1 MB)

Hyperspectral sensors collect hundred or even thousands of spectral bands at different wavelengths of the same area on the Earth surface, which generates large data volumes comprising several GBs per flight. This high spectral resolution can be used for object detection and for discrimination between different objects based on their spectral characteristics.

In recent years, graphics processing units (GPUs) have evolved into highly parallel and programmable systems. Specifically, several hyperspectral imaging algorithms have shown to be able to benefit from this hardware taking advantage of the extremely high floating-point processing performance, compact size, huge memory bandwidth, and relatively low cost of these units, which make them appealing for onboard data processing.

This paper proposes an efficient implementation of SISAL algorithm for unsupervised hyperspectral linear unmixing on GPUs using CUDA. SISAL aims to identify the endmembers of a scene, {em i.e.}, is able to unmix hyperspectral data sets in which the pure pixel assumption is violated. The parallel implementation of SISAL presented in this work exploits the GPU architecture at low level, using shared memory and coalesced accesses to memory. The experimental results herein presented indicate that the GPU implementation can significantly accelerate the execution of SISAL over big datasets, when compared with the sequential version which has been carefully optimized for one CPU core.