HYPERMIXUna nueva herramienta para desmezclado de imágnes hiperespectrales en GPUs

Abstract

Spectral unmixing is an important problem for remotely sensed hyperspectral image exploitation, as most of these images have limited spatial resolution that results in the presence of several pure components (called endmembers) in each pixel of the scene. As a result, spectral unmixing is important to express each pixel as a collection of endmembers weighted by their corresponding proportions (called abundances), thus allowing an interpretation of the images at a sub-pixel level. In this regard, the newly developed tool (called HyperMix) offers a state-of-the art development in the field, as it contains computationally efficient implementations of algorithms that can cover all the different stages involved in the hyperspectral unmixing chain, which aims to provide (from the original hyperspectral image) a set of endmembers and their corresponding abundance fractions in each pixel in unsupervised fashion. One important contribution of the HyperMix tool is the fact that most of the algorithms that it contains are implemented in computationally efficient fashion. For this purpose, we exploit commodity graphics processing units (GPUs) as a hardware resource that provides high performance computing at very low cost. Specifically, our implementations are developed using the Compute Unified Device Architecture (CUDA) from NVidia.