High-level FPGA-based implementation of a hyperspectral endmember extraction algorithm

Linear spectral unmixing represents an awesome technique for the analysis of remotely sensed hyperspectral images. However, its large computational cost severely compromises its use in applications under real-time constraints, where swift responses are of a crucial importance. Hence, the hardware acceleration of the operations involved in the unmixing of a hyperspectral cube becomes mandatory for these scenarios. This paper presents an improved version of a design flow that allows implementing a hyperspectral unmixing algorithm onto a Field Programmable Gate Array (FPGA) directly from MATLAB. As a case of study, the results obtained with the implementation of the well-known N-FINDR algorithm will be outlined, demonstrating the benefits of our proposal against state-of-the-art approaches as well as the profits derived from the adoption of fixed rather floating-point arithmetic. The presented high level methodology can be easily extrapolated to the implementation of other hyperspectral MATLAB algorithms, drastically accelerating the design cycle from concept to implementation.