High-level synthesis acceleration for an FPGA implementation of an optimized automatic target detection and classification algorithm for hyperspectral image analysis with Intel oneAPI

For many years, hyperspectral images have been one of the most well-established technologies in remote sensing. The main difficulty in hyperspectral image analysis lies in the spectral mixing of materials within a single pixel, which hinders the identification of pure components or endmembers. This task is essential for various Earth observation applications, such as agriculture, mining, and environmental management. From the whole process, the endmember identification or target detection is usually one of the most time-consuming stages, so high-performance computing (HPC) platforms such as multicore processors, graphics processing units (GPUs) or field-programmable gate arrays (FPGAs) are necessary and essential for its exploitation in time-critical scenarios. In this article, we present a high-level synthesis (HLS) acceleration for an FPGA implementation of the automatic target detection and classification algorithm (ATDCA) using the Gram–Schmidt (GS) method for hyperspectral images with the Intel oneAPI Toolkit and DPC++ instead of traditional hardware description languages (HDL). Optimization strategies were applied in terms of parallelism and efficiency in the use of hardware resources on a Stratix 10 SX 2800 FPGA, resulting in a significant performance improvement. Experimental results showed that the optimized implementation through HLS achieved a significant reduction in processing times, demonstrating that the use of optimization techniques for FPGA platforms, combined with the DPC++ environment, provides an effective and flexible solution for spectral unmixing of hyperspectral images.