On the use of Jetson TX1 board for parallel hyperspectral compressive sensing

Hyperspectral imaging instruments measure hundreds of spectral bands (at different wavelength channels) for the same area of the surface of the Earth. Typically the data cube collected by these sensors comprises several GBs per flight, which have attracted attention to on-board techniques for compression. Typically these compression techniques are expensive from the computational point of view. Due to this fact, a number of Compressive Sensing and Random Projection techniques have raised as an alternative to reduce the signal size on-board the sensor. The measuring process of these techniques usually consist on performing dot products between the signal and random vectors. The Compressive Sensing process is performed directly in the optic system, however, in this paper, we propose to perform the random projection measurement process on a low power consumption Graphic Processing Unit. The experiments are conducted on a Jetson TX1 board, which is well suited to perform vector operations such as dot products. These experiments have been performed to demonstrate the applicability, in terms of accuracy and time consuming, of these methods for onboard processing. The results show that by using this low power consumption GPU is it possible to obtain real-time performance with a very limited power requirement.