Active galactic nuclei (AGN) play a key role in galaxy evolution but are challenging to identify due to their varied observational signatures. Furthermore, understanding their impact requires quantifying their strength relative to their host galaxies. We developed a deep learning (DL) model for identifying AGN in imaging data by deriving the contribution of the central point source. Trained on Euclidised mock galaxy images with injected AGN levels, in the form of varying contributions of the point-spread function (PSF), our model can precisely and accurately recover the injected AGN contribution fraction f ₒ₅, with a mean difference between the predicted and true f ₒ₅ of -0. 0078 and an overall root mean square error (RMSE) of 0. 051. This method moves beyond binary AGN classification, enabling precise AGN contribution measurements. Applying our model to a stellar-mass-limited sample (M_ 10^9. 8 M_, 0. 5 z 2. 0) from the first quick data release (Q1), we identify 48, 840 78 AGN over 63. 1 deg² (7. 80. 1%) using a threshold of f ₒ₅ > 0. 2. We compare our DL-selected AGN with those identified in X-ray, mid-infrared (MIR), and optical spectroscopy and investigate their overlapping fractions depending on different thresholds on the PSF contribution. We find that the overlap increases with increasing X-ray or bolometric AGN luminosity. The AGN luminosity in the I ₄ filter correlates with host galaxy stellar mass, suggesting faster supermassive black hole (SMBH) growth in more massive galaxies. Moreover, the mean relative contribution of the AGN is higher in quiescent galaxies than in star-forming ones. Starburst galaxies and the most massive galaxies (across the star-formation main sequence) tend to host the most luminous AGN, indicating concomitant assembly of the SMBH and the host galaxy.
Serrano et al. (Wed,) studied this question.