Multi-frequency point source detection with fully convolutional networks: Performance in realistic microwave sky simulations
Point Source (PS) detection is an important issue for future Cosmic Microwave Background (CMB) experiments since they are one of the main contaminants to the recovery of CMB signal at small scales. Improving its multifrequency detection would allow to take into account valuable information otherwise neglected when extracting PS using a channel-by-channel approach. We develop a method based on Neural Networks (NNs) to detect PS in multifrequency realistic simulations and compare its performance against one of the most popular methods, the matrix filters. The frequencies used are 143, 217 and 353 GHz and we impose a Galactic cut of 30 degrees. We produce simulations by adding contaminating signals to the PS maps as the CMB, the Cosmic Infrared Background, the Galactic thermal emission, the thermal Sunyaev-Zel'dovich effect and the instrumental noise. These simulations are used to train two NNs called Flat and Spectral MultiPoSeIDoN. The first one considers PS with a flat spectrum and the second one is more realistic because it takes into account the spectral behavior of the PS. Using a detection limit of 60 mJy, Flat MultiPoSeIDoN reachs the 90% of completeness level at 58 mJy and at 79, 71 and 60 for the spectral case at 143, 217 and 353 GHz respectively, while the matrix filters reach it at 84, 79 and 123 mJy. Using safer 4σ detection limit does not help to improve these results. In all cases, MultiPoSeIDoN obtain a much lower number of spurious sources than the filter. The NNs recover the flux density of the detections with a relative error of 10% above 100 mJy, while the filter above 150 mJy. Based on the results, NNs are the perfect candidates to substitute filters to detect multifrequency PS in future CMB experiments. Moreover, we have shown that a multifrequency approach can detect sources with higher accuracy than single-frequency approaches also based on NNs.