A Hardware Accelerator for the Segmentation of Hyperspectral Images

摘要

Hyperspectral images (HSIs) contain hundreds of spectral bands that combine spatial characteristics of a scene and are widely used in object identification and classification in various industry and science contexts. However, the processing of HSIs requires high processing power, and one of its compute-intensive tasks is the image segmentation. In this context, this paper presents the development of a first hardware accelerator proposed to speedup a multiresolution algorithm for the segmentation of HSIs. The most computationally expensive step of this algorithm was described in HDL and synthesized to FPGA. Results show that the performance of the hardware accelerator is close to that of a 3.2 GHz PC desktop, spending 20% of the energy consumed by the PC to process a single band.