Iterative Normalized Matched Filtering for Detection of Chemical Agents in Hyperspectral Imaging

Abstract

Hyperspectral imaging (HSI) can be used to detect a harmful chemical agents' (CAs,) cloud from a long distance. A normalized matched filter (NMF) is one of the best algorithms to detect CAs in the atmosphere with perfectly known statistics of the background. However, if the statistics of the background are affected by a CA's signal, (that is a contamination condition) the performance of the NMF detector is degraded. To design an NMF detector that is robust to contamination, we propose an iterative normalized matched filter (INMF). The proposed algorithm extracts CA-off spectra from the contaminated background spectra dataset using a contaminated NMF detector. And the NMF detector is designed using the extracted CA-off background spectra and this procedure repeats until convergence. Simulation results demonstrate that the proposed algorithm significantly improves the detection performance.