GPUs versus FPGAs for Onboard Payload Compression of Remotely Sensed Hyperspectral Data

In this paper, we compare field programmable gate arrays (FPGAs) versus graphical processing units (GPUs) in the framework of (lossy) remotely sensed hyperspectral data compression by developing parallel implementations of a spectral unmixing-based compression strategy on both platforms. For the FPGA implementations, we resort to Xilinx hardware devices certified for on-board operation, while for the GPU implementation we make use of hardware devices available from NVidia, such as the Tesla series. In our comparison, we thoroughly assess the advantages and disadvantages of each considered architecture by designing and evaluating new parallel compression algorithms for both types of platforms. These algorithms are quantitatively evaluated in terms of lossy compression performance using hyperspectral data collected by the NASA's Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) system over the World Trade Center (WTC) in New York, five days after the terrorist attacks that collapsed the two main towers in the WTC complex. Our experimental results indicate that low-weight and low-power integrated components are very appealing to reduce mission payload and obtain analysis/compression results in real-time, thus bridging the gap towards on-board lossy compression of remotely sensed hyperspectral data.