PROJECT TITLE :

Parallel Hyperspectral Unmixing Method via Split Augmented Lagrangian on GPU

ABSTRACT:

One in every of the main issues of hyperspectral information analysis is the presence of mixed pixels due to the low spatial resolution of such pictures. Linear spectral unmixing aims at inferring pure spectral signatures and their fractions at every pixel of the scene. The huge information volumes acquired by hyperspectral sensors put stringent necessities on processing and unmixing ways. This letter proposes an economical implementation of the tactic called simplex identification via split augmented Lagrangian (SISAL) that exploits the graphics processing unit (GPU) design at low level using Compute Unified Device Design. SISAL aims to spot the endmembers of a scene, i.e., is in a position to unmix hyperspectral information sets in that the pure pixel assumption is violated. The proposed implementation is performed in an exceedingly pixel-by-pixel fashion using coalesced accesses to memory and exploiting shared memory to store temporary knowledge. Furthermore, the kernels have been optimized to reduce the threads divergence, thus achieving high GPU occupancy. The experimental results obtained for the simulated and real hyperspectral data sets reveal speedups up to forty nine times, that demonstrates that the GPU implementation can significantly accelerate the method's execution over huge knowledge sets whereas maintaining the ways accuracy.