Scalable architectures for real-time hyperspectral unmixing

Abstract

Hyperspectral imaging instruments capture and collect hundreds of different wavelength data corresponding to the same surface. As a result, tons of information must be stored, processed and transmitted to ground. However, the downlink bandwidth is limited, and transmitting all data from the satellite to ground is a slow task that jeopardizes the use of this information for applications under real-time or near real-time constraints. This is the reason why most of the research activity is moving towards developing solutions which are able to process this data on-board, sending back to ground only relevant information.

In this context, this paper presents three different FPGA-based architectures, implemented on an FPGA, which perform the Modified Vertex Component Analysis (MVCA) algorithm, as part of the hyperspectral linear unmixing processing chain. As a consequence, not only high performance but flexible and reusable designs have been designed. Moreover, final results demonstrate the influence of the implemented parallelization methodology over the final performance which motivates the system adaptivity through the scalability feature.