Chinese scientists created ASTERIS AI to boost deep-space imaging, revealing faint galaxies and advancing astronomy.

Chinese scientists from Tsinghua University have developed an AI model named ASTERIS that enhances deep-space imaging by reducing noise and improving detection of faint celestial objects. Using self-supervised spatiotemporal denoising and a photometric adaptive screening mechanism, the model processes telescope data as 3D volumes, increasing sensitivity by 1.0 magnitude and enabling the discovery of over 160 high-redshift galaxy candidates from the early universe—tripling prior rates. Applied to James Webb Space Telescope data, it extended observations into mid-infrared wavelengths and revealed objects 2.5 times fainter than before. The technology is compatible with multiple telescopes and could advance research on dark energy, dark matter, and exoplanets, with potential use in next-generation observatories. The findings were published in Science.