SO Webloquio: When artificial intelligence meets astronomy: celestial object census

Over the centuries, astronomers have continued to improve the performance of telescopes and the techniques for observing and analysing data. Nowadays, humans are building more and more advanced telescopes with larger and deeper observations, reaching terabytes and even petabytes of data. The Square Kilometre Array (SKA) radio telescope, the most ambitious project in astronomy under construction, is expected to produce more than 700 petabytes of scientific data per year since 2029. One of the main observational tasks of the super-telescope SKA is to conduct a 'celestial object census'. This is a huge work! For example, the SKA pathfinders’ survey projects are expected to detect 70 million radio galaxies. The classification and morphology of these radio sources provides key information for understanding the formation and evolution of the Universe. However, in the big data era, the challenge for astronomers worldwide has become how to access and use this vast amount of information. It is clearly impossible to identify and classify the countless number of celestial objects through visual inspection. To meet the challenge of the data deluge, astronomers seek automated and intelligent methods of data processing using supercomputers. When artificial intelligence meets astronomy new opportunities are created for scientific breakthroughs. The SKA team from Shanghai Astronomical Observatory has used artificial intelligence to develop a source finding tool and named it ‘HeTu’ (meaning Sky Map in traditional Chinese myths and legends). HeTu not only increases the depth of the deep learning network and improves accuracy, but also provides a map of the features of multi-scale objects. Experiments have shown that HeTu is able to rapidly locate, identify and classify both compact sources and extended radio sources in an automated manner, and that the results of HeTu are not dependent on the dataset used, making it more widely adaptable.