On February 20, according to Tsinghua University, the Imaging and Intelligent Technology Laboratory led by Academician Dai Qionghai from the Department of Automation, in collaboration with Associate Professor Cai Zheng from the Department of Astronomy, developed the AI astronomical observation enhancement model “ASTERIS” (Xingyan). The model successfully surpassed the depth limits of astronomical observation, significantly enhancing the detection capability of the James Webb Space Telescope (JWST). The related research results were published in the international academic journal Science.
Currently, traditional astronomical observation relies heavily on hardware upgrades, which have reached marginal gains bottlenecks, and complex spatiotemporal heterogeneous noise further complicates the detection of extremely faint celestial objects. The team’s ASTERIS model innovatively implements a photometric adaptive screening mechanism, jointly modeling noise and celestial luminosity, and employs a “time-segment median, full-time averaging” optimization strategy to remove transient interference and enhance the signal-to-noise ratio (SNR) of faint signals. This approach ensures high-fidelity signal reconstruction while strictly maintaining the scientific rigor of astronomical data.
Experimental data show that ASTERIS increases the JWST detection depth by 1 magnitude, improves photon collection efficiency by nearly an order of magnitude, and effectively increases the equivalent observational aperture from 6.4 meters to nearly 10 meters. Using this model, the team identified over 160 high-redshift candidate objects dating from 200 million to 500 million years after the Big Bang, which is three times the number identified in previous studies, producing the deepest and most detailed images of extreme deep-space galaxies to date. These results provide crucial new data for studying the origin of galaxies during the cosmic dawn.
ASTERIS also exhibits strong generalization capabilities, adapting to multiple telescopes and multi-band observations without manual labeling, and has been successfully applied to both space-based and ground-based astronomical observation instruments. This achievement represents a shift from hardware-centric observation to AI-enhanced astronomical observation, providing core technological support for humanity’s exploration of the origin of the universe and other frontier scientific problems.
Source: Hua Ling, Science and Technology Daily