中山大学Napolitano教授团队开发出基于机器学习的星系结构参数拟合工具GaLNets，可以短时间获取千万个星系的参数

The China Space Telescope (CSST) will observe billions of galaxies with unprecedented accuracy providing us images with details similar to the Hubble Space Telescope. With this high-quality imaging, we have a unique chance to study the evolution of galaxies up to the early epochs of their formation. We will be able to understand the mechanisms that have transformed galaxies, across time, from spiral, disk-dominated systems, to elliptical galaxies. In particular, we will investigate the role that lenticular galaxies, the intermediate class between them, have played in this evolution. To do this we need to be able to measure the physical parameters like their luminosities, colours, sizes, ellipticities of the disks and the inner spheroidal component (bulges) separately of up to a billion of galaxies in the CSST data. This will be possible only using machine learning techniques that can learn how to predict the galaxy parameters by extracting from galaxy images the information about the way the light is distributed within them from their 2-dimensional surface brightness distribution. A collaboration between Sun Yat-sen University and the National Observatory of China is developing a series of Convolutional Neural Networks that have been called GAlaxy Light profile convolutional neural NETworks (GaLNets), that can produce these "structural parameters" for the CSST observations in a fast and accurate way, by processing a billion of galaxies in only a few days, a task that with standard analyses would take months. The GaLNets are currently being successfully applied to ground-based observations like the Kilo Degree Survey (KiDS).