From DES to KiDS: Domain adaptation for cross-survey detections of low-surface-brightness galaxies

Abstract

Context . Low-surface-brightness galaxies (LSBGs) are vital for understanding galaxy formation, but their diffuse nature makes them challenging to detect. Upcoming large-scale surveys are expected to uncover a large number of LSBGs, which require robust automated methods to identify them across heterogeneous datasets. As a precursor to the Legacy Survey of Space and Time (LSST) and Euclid , we explore domain adaptation techniques to build homogeneous LSBG catalogues across current surveys. Aims . We investigate the use of computer vision models and domain adaptation for cross-survey LSBG identification. Using models trained on Dark Energy Survey (DES) data, we search for LSBGs in the Kilo-Degree Survey Data Release 5 (KiDS DR5). We examine their structural and stellar population properties to pave the way for large-scale LSBG studies with LSST and Euclid . Methods . We used an ensemble consisting of one convolutional neural network (CNN) and two transformer models trained on DES cutouts and applied to KiDS DR5 imaging with surface-brightness normalisation. Structural parameters were estimated with galfitm and the sample was further refined through visual inspection to produce the final candidate sample. Photometric redshift and stellar population properties were estimated through spectral energy distribution (SED) fitting with CIGALE. Results . We identified 20 180 LSBGs and 434 ultra-diffuse galaxies (UDGs) in KiDS DR5. Their structural parameters are similar to the known LSBGs from DES and Hyper Suprime-Cam SSP Survey (HSC-SSP). The KiDS-LSBGs follow a continuous size–luminosity relation connecting classical dwarf galaxies and UDGs and their colours are bimodal (~73% blue, ~27% red). Crossmatching with spectroscopic and cluster catalogues provided redshifts for 4913 systems, enabling a systematic characterisation of the star-forming main sequence of LSBGs. Photometric redshifts derived via SED fitting are mildly overestimated (≃0.024), leading to systematic offsets in stellar mass and star formation rate estimates. However, these biases induce only small shifts (~0.13–0.22 dex) in terms of the specific star formation rate, thereby preserving the structure of the star-forming main sequence. Strong environmental trends are also evident, with cluster LSBGs and UDGs exhibiting redder colours and reduced star formation compared to non-cluster systems. This indicates efficient quenching driven by environmental processes. Conclusions . We demonstrate that with domain adaptation, cross-survey LSBG identification can be achieved with deep learning models. Thus, the methodology presented here provides a powerful and scalable pathway for constructing homogeneous LSBG catalogues across surveys. This framework is well-suited for the era of LSST and Euclid , when millions of diffuse galaxies will be discovered and consistent cross-survey classifications will become essential.