Abstract
Multispectral, Imaging, Detection, and Active Reflectance (MiDAR) is a method capable of imaging through ocean waves without distortion in 3D at sub-cm resolutions to sense living and non-living structures in light-limited and analog planetary science environments. MiDAR utilizes high intensity narrowband structured optical radiation to illuminate a target and measure the object's spectral reflectance at fine and spatial temporal scales with a signal-to-noise ratio 10-10 3 higher than passive airborne and spaceborne remote sensing systems. The MiDAR data is then reconstructed using the computational imaging capability associated with airborne platform such as unmanned aerial vehicles (UAVs) and spaceborne platforms such as a CubeSat. UAVs deploy embedded Graphics Processing Unit (GPU) such as Nvidia's Tegra, while CubeSat consists of Field Programmable Gate Array (FPGA) devices such as Xilinx's MPSoC. The challenges involved in mapping MiDAR to each platform is unique and this paper describes the implementation and optimization steps to achieve maximum performance from each accelerator using CUDA and High Level Synthesis (HLS). The FPGA implementation using HLS is 14.7x faster than the CUDA-based GPU implementation and also provides a higher throughput of 13,552 fps using Xilinx MPSoC as compared to 921 fps achieved using Nvidia Tegra.