Prediction of stellar mass and star formation rate for low-redshift galaxies in the DESI Legacy Imaging Surveys using deep learning

We present an extensive catalog of the stellar mass (M*) and specific star formation rate (sSFR) for about 18 million galaxies with low-redshift (z<0.3) observed by the Dark Energy Spectroscopic Instrument (DESI) Legacy Imaging Surveys. Using a deep learning framework based on EfficientNet (GalEffNet), we extract features from photometric images to predict M* and sSFR. The testing results demonstrate that our predictions of M* and sSFR are in good agreement with traditional spectroscopic estimates, with 1σ scatters of 0.221 dex for M* and 0.411 dex for sSFR. We systematically analyze the predictions across four morphological types-DEV, EXP, REX, and SER-evaluating model performance per type. Applying our deep learning technique to the DESI Legacy Imaging Surveys Data Release 9, we provide a catalog that encompasses the extensive estimates of M* and sSFR for approximately 18 million galaxies. Using this catalog, we present the sSFR vs. M* diagram, in which the distinct distributions of different morphological types highlight the role of galaxy structure in understanding galaxy evolution.