アブストラクト：
We search for extremely metal-poor emission-line galaxies (EMPGs) lurking in the local universe (z < 0.03). EMPGs show primordial properties such as very low metallicity ( < 10% Zsun), low stellar mass (10^5-7 Msun), young star population ( < 30 Myr), and high-ionization. The EMPG is thus a key galaxy population helpful to understand the early-epoch galaxy-formation/evolution and the cosmic reionization. In this study, we use a deep, wide-field imaging data of Subaru/HSC-SSP to search for EMPGs, which are very rare in the local universe. Thanks to the HSC data, 5-7 mag deeper than SDSS data, we can discover very faint EMPGs that previous studies cannot find out in the SDSS data. To select EMPG candidates effectively, we have developed a selection technique based on deep learning. In principal, we distinguish EMPGs with broad-band excesses caused by strong nebular lines (Balmer lines and [OIII]) + nebular continuum that are unique to EMPGs. We apply our selection technique to the HSC-SSP data, obtaining about ten EMPG candidates. In our poster, we introduce our selection technique and discuss the nature of EMPGs (very low metallicity, young stellar population, hard ionizing radiation etc.) with a report of our spectroscopic observations in progress.