Title:Efficient Search for Extremely Metal Poor Galaxies in the Local Universe using Convolutional Neural Networks

Abstract:Nearby extremely metal-poor galaxies (XMPs) allow us to study primitive galaxy formation and evolution in greater detail than is possible at high redshift. This work, for the first time, promotes the use of convolutional neural networks (CNNs) to efficiently search for XMPs in multi-band imaging data based on their predicted N2 index (N2\,$\equiv\log$\{\rNii/\Ha\}). We developed a sequential characterisation pipeline, composed of three CNN procedures: (i) a classifier for metal-poor galaxies, (ii) a classifier for XMPs, and (iii) an N2 predictor. The pipeline is applied to over 7.7 million SDSS DR17 imaging data without SDSS spectroscopy. The predicted N2 values are used to select promising candidates for observations. This approach was validated by new observations of 45 candidates with redshifts less than 0.065 using the 2.54~m Isaac Newton Telescope (INT) and the 4.1~m Southern Astrophysical Research (SOAR) Telescope between 2023 and 2024. All 45 candidates are confirmed to be metal-poor, including 28 new discoveries. There are 18/45 galaxies lacking detectable \rNii\ lines ($S/N<2$); for these, we report $2\sigma$ upper limits on their oxygen abundance. Our XMPs have estimated oxygen abundances of $7.1\leq$\OH$\leq8.7$ ($2\sigma$ upper limit), based on the N2 index, and 21 of them with estimated metallicity $<0.1~Z_{\odot}$. Additionally, we identified 4 potential candidates of low-metallicity AGNs at $\lesssim0.1Z_{\odot}$. Finally, we found that our observed samples are mostly brighter in the $g-$band compared to other filters, similar to blueberry (BB) galaxies, resembling green pea galaxies and high-redshift Ly$\alpha$ emitters.