Finding sources, such as galaxies or stars, is a classical issue of astronomical surveys. Many methods of sources detection are available in the literature. The detection method based on a marked point process in a Bayesian framework proposed in our previous work [1,2] has demonstrated its ability to meet the challenge proposed by faint galaxies detection in hyperspectral MUSE data. However to detect the faintest galaxies, we must accept the presence of false detections. Recent studies demonstrate the importance of introducing statistical methods to characterize the detection, for instance by giving a false alarms rate or a false discoveries rate (FDR) control procedure. The limitation of our detection method is twofold: the computational complexity and the error control for the detection. To partially solve these two issues, we propose in this work a preprocessing step which reduces the exploration of the configuration space and which introduces a pixel wise type I error control. Different approaches based on multiple testing are presented. 

[1] F. Chatelain, A. Costard, and O. Michel, “A Bayesian marked point process for object detection. Application to MUSE hyperspectral data,” in International Conference on Acoustics, Speech and Signal Processing, 2011. 

[2] C. Meillier, F. Chatelain, O. Michel, and H. Ayasso, “Non-parametric Bayesian framework for detection of object configurations with large intensity dynamics in highly noisy hyperspectral data,” in International Conference on Acoustics, Speech and Signal Processing, 2014.