AutoSourceID-FeatureExtractor. Optical images analysis using a Two-Step MVE Network for feature estimation and uncertainty characterization

Aims. In astronomy, machine learning has demonstrated success in various tasks such as source localization, classification, anomaly detection, and segmentation. However, feature regression remains an area with room for improvement. We aim to design a network that can accurately estimate sources' features and their uncertainties from single-band image cutouts, given the approximated locations of the sources provided by the previously developed code ASID-L or other external catalogues. Methods. The algorithm presented here, AutoSourceID-FeatureExtractor (ASID-FE), uses single-band cutouts of 32x32 pixels around the localized sources to estimate flux, sub-pixel centre coordinates, and their uncertainties. ASID-FE employs what we call a TS-MVE, a Two-Step Mean Variance Estimator approach to first estimate the features and then their uncertainties without the need for additional information, e.g. Point Spread Function (PSF). Results. We show that ASID-FE, trained on synthetic images from the MeerLICHT telescope, can predict more accurate features with respect to similar codes like SourceExtractor and that the two-step method can estimate well-calibrated uncertainties that are better behaved compared to similar methods that use deep ensembles of simple MVE networks. Finally, we evaluate the model on real images from the MeerLICHT telescope and the Zwicky Transients Facility (ZTF) to test its Transfer Learning abilities.