Abstract:

The automatic target generation process (ATGP) is unsuitable for being deployed for high speed hardware implementation due to enormously complex inverse operations and the increasing scale of iterative operation problems. On the basis of an intensive study of the ATGP algorithm, a novel implementation framework is proposed in this paper, which takes advantage of simple regular matrix operations instead of increasingly complicated matrix inversion to update the orthogonal projection operator matrix. Furthermore, an accelerated design of the algorithm is implemented on FPGA by using high-level synthesis (HLS) for the first time. Experimental results demonstrate that the processing time achieved by the FPGA implementation is strictly real-time while retaining the same high detection accuracy.

Key words: image processing, hyperspectral image, endmember extraction, automatic target generation process, field programable gate array, high-level synthesis