Accelerating the ATDCA Algorithm for Endmember Extraction from Hyperspectral Imagery with Intel oneAPI for FPGAs

The decomposition of the observed pixel spectrum of hyperspectral data into its constituent spectral signatures (or endmembers) and a set of corresponding fractional abundances is known as hyperspectral unmixing. To carry out this analysis, it is required to use a computational technology with high-performance computing due to the high dimensionality of the data and the high complexity of the unmixing algorithms. Field Programmable Gate Arrays (FPGAs) are excellently suited to processing needs since they provide flexibility, low power consumption, and great performance. Faster alternative development methods are needed because the rate at which new algorithms for hyperspectral analysis are developed is substantially higher than the rate at which these algorithms are implemented on FPGAs. High-level synthesis (HLS) is a technology that assists with the transformation of a behavioral description of hardware into a register transfer level (RTL) model, thus reducing development times. In this work, the automatic target classification and detection algorithm (ATDCA) has been accelerated on FPGAs using Dataparalell C++ (DPC++) and Intel oneAPI. The analysis of the algorithm and the three accelerations carried out achieve a speedup above 13×.