On the Evaluation of Machine Learning Algorithms for Hyperspectral Image Classification on a Heterogeneous Computing Device

Abstract

Hyperspectral imaging (HSI) systems collect electromagnetic radiation, emitted or reflected by the scene under observation at, typically, hundreds of contiguous and regularly spaced narrow spectral bands, from the visible to the infra-red region of the spectrum. The result is a 3D data cube with two spatial dimensions and a wavelength dimension, composed of hundreds of co-registered images containing information on the spectral signature of the materials present in the field of view of the sensor at the time of capture. Retrieving this information in an efficient way, requires compute systems capable of dealing with a huge amount of data and complex algorithms. The recent development of single board computers (SBC) with enough compute power to deal with HSI processing using machine learning (ML) algorithms opens new opportunities for embedded HSI systems for Earth Observation remote sensing, laboratory setups and industrial applications. In this work, we evaluate the performance of the ODROID-XU4, a heterogeneous computing device, running ML algorithms for HSI classification. The experimental results show the potential of SBC as alternative hardware platforms for applications requiring low power and low cost devices, such as onboard processing units and edge computing.