ABSTRACT
This chapter describes the computationally efficient implementations of two different approaches for Compressive sensing (CS) and Random Projections (RP) on graphics processing units (GPUs): hyperspectral coded aperture (HYCA) algorithm for CS and spectral compressive acquisition (SpeCA) as a form of RP technique. It shows the implementation of these algorithms with several optimizations for accelerating their computational performance while maintaining their accuracy. Our study reveals that the implementations on GPU can provide real-time performance. VCA endmember extraction algorithm to find matrix, parameterized with endmembers is considered. The proposed implementations for HYCA and SpeCA are detailed. The experimental results are reported to indicate that remotely sensed hyperspectral imaging can greatly benefit from the development of efficient implementations of CS algorithms in specialized hardware devices for better exploitation of high-dimensional data sets.
