Chinese researchers from Tsinghua University have developed a new artificial intelligence model that substantially improves astronomers' ability to study deep space.
The model is called ASTERIS — Astronomical Spatiotemporal Enhancement and Reconstruction for Image Synthesis. It combines computational optics methods and AI to capture extremely faint signals from distant objects.
The results of the work have been published in the journal Science. By applying ASTERIS to data from the James Webb telescope, the scientists were able to extend the observation range from visible light at about 500 nanometers to mid-infrared radiation at 5 micrometers.
The detection depth increased by one magnitude: the telescope now records objects 2,5 times fainter than before. This made it possible to identify more than 160 candidate galaxies with high redshift from the era of "Cosmic Dawn" — the period 200–500 million years after the Big Bang.
This number is three times greater than the results of previous methods. According to Associate Professor Cai Zheng of the Department of Astronomy, the model helps overcome sky noise and the thermal radiation of telescopes, which previously masked faint signals.
ASTERIS uses a self-supervised spatiotemporal denoising method and constructs three-dimensional images, accounting for changes in time and space. An additional adaptive photometric selection mechanism separates real signals from background noise.
Professor Dai Qionghai of Tsinghua's Department of Automation noted that hidden objects can now be reconstructed clearly and accurately. The technology works with different observation systems and could become a universal platform for processing large volumes of data.
In the future, the model is planned to be used in next-generation telescopes. How will this affect the search for answers about dark energy, dark matter and the origin of the Universe?
The scientists emphasize that ASTERIS is already demonstrating potential in the detection of exoplanets and other key tasks of modern astronomy.
