PDR and LRMAP detection tests applied to massive hyperspectral data

Recent works showed that two composite detection tests based on Maximum A Posteriori (MAP) estimates can be more powerful than the Generalized Likelihood Ratio (GLR) in the case of sparse parameters. These tests are the Posterior Density Ratio (PDR), which computes the ratio of the a posteriori distribution under each hypothesis, and the LRMAP, where the MAP replaces the Maximum Likelihood estimate. We propose here a compared analysis of the two MAP-based tests performances. The implementation details of these tests are then analyzed in the framework of massive hyperspectral data which will be acquired by the MUSE (Multi Unit Spectroscopic Explorer) integral field spectrograph. We finally improve the detection strategy proposed in [8] by better exploiting the spatial dependencies existing in the data cube.