Subpixel Target Detection in Hyperspectral Imaging

Detecting a target of known spectral signature from an unknown background is one of main goal of hyperspectral imaging. As the majority of hyperspectral imaging systems have a poor spatial resolution, subpixel targets are usual. In this case, the so-called replacement model is commonly advocated. This model, valid for reflectance images, specifies that if a target is present, the amount of background should reduce in the same proportion. Nevertheless, the majority of the standard detectors, such as the Match Filter or the Kelly detector, have been developed for different contexts, and do not exploit this constraint. One of the rare example that is suitable for the replacement model is the Finite Target Match Filter, which is known to improve the target selectivity detection. In this paper, we develop the exact Generalized Likelihood Ratio Test for the model at hand. We show that this new detector outperforms the standard ones, on a real data experiment.