Harnessing AI to map the cosmos, Japanese scientists uncover hidden structures in the Milky Way—revealing fresh insights into star formation and explosive galactic events.

The images on the left show newly detected bubble-like structures, while the images on the right show the bubble-like structure detected in this study and previous studies. By using wavelengths of 8 μm (green) and 24 μm (red), it is possible to detect the bubble structures created by the formation of high-mass stars. Image Credit: Osaka Metropolitan University
Japanese researchers have developed a deep-learning model to learn more about the farthest reaches of our galaxy and the mysteries surrounding star formation. The team from Osaka Metropolitan University utilized artificial intelligence to analyze the vast amounts of data collected from space telescopes, discovering bubble-like structures not previously included in existing astronomical databases.
Like other galaxies in the universe, the Milky Way features bubble-like structures mainly formed during the birth and activity of high-mass stars. These so-called Spitzer bubbles provide important clues for understanding the processes of star formation and galaxy evolution.
Graduate School of Science student Shimpei Nishimoto and Professor Toshikazu Onishi collaborated with scientists across Japan to create the deep learning model. Utilizing data from the Spitzer Space Telescope and the James Webb Space Telescope, the model employs AI image recognition to accurately and efficiently detect Spitzer bubbles. Additionally, they identified shell-like structures believed to have formed from supernova explosions.
"Our findings indicate that it is possible to conduct detailed investigations not only of star formation but also of the effects of explosive events within galaxies," said graduate student Nishimoto.
Professor Onishi added, "In the future, we hope that advancements in AI technology will accelerate the understanding of the mechanisms behind galaxy evolution and star formation."
Source:
Journal reference: