A reconfigurable multi-mode implementation of hyperspectral target detection algorithms
Peer reviewed, Journal article
Accepted version
Åpne
Permanent lenke
https://hdl.handle.net/11250/2723440Utgivelsesdato
2020Metadata
Vis full innførselSamlinger
Originalversjon
https://doi.org/10.1016/j.micpro.2020.103258Sammendrag
Hyperspectral images obtained by imaging spectrometer contain a large data amount that requires techniques such as target detection for information extraction. The proposed multi-mode FPGA implementation combines matrix correlation and inversion matrix computations by using the Sherman-Morrison method to achieve real-time operation. The implementation supports Constrained Energy Minimization (CEM), Adjusted Spectral Matched Filter (ASMF) and modified Adaptive Cosine Estimator (ACE) detectors. The detection performance of the algorithms is evaluated using standard detection metrics. The proposed implementation has been realized on Zynq family SoCs and verified against the MATLAB reference software. The detection results for different fixed-point data types and target detection algorithms are reported. Finally, the proposed implementation is compared with state-of-the-art designs in terms of both throughput and resource utilization.