A research team led by Prof. LONG Qian from the Lijiang Observatory of Yunnan Observatories of the Chinese Academy of Sciences has constructed a class of two-channel deep convolutional neural network model to search for clusters in the Andromeda Galaxy.

The study was published in Astronomy & Astrophysics on Feb. 1.

Dr. WANG Shoucheng and Prof. MA Jun from the National Astronomical Observatories and Associate Professor CHEN Bingqiu from Yunnan University were also involved in the study.

Clusters are widely distributed throughout the galaxy from the nuclear sphere and disk to the outer halo, documenting the early formation and evolutionary history of galaxies.

The Andromeda Galaxy M31 is the closest large spiral galaxy to our Milky Way and is an ideal astrophysical laboratory for astronomers to study galaxy formation and evolution.

Recent wide-field photometric and spectroscopic surveys have provided an opportunity to identify the M31 clusters. But to find and identify the special objects from tens of millions of images provided by the deep wide-field photometric surveys is difficult at present.

In the study, the researchers selected the M31 clusters, Galactic foreground objects and background galaxies from the LAMOST DR6 data, and combined the M31 clusters and non-cluster samples given in the literature as training samples. The proposed model achieved an accuracy of 99% in the test set. 

Using this model, the researchers identified 117 high-confidence M31 cluster candidates from more than 21 million images obtained by the PAndAS photometric survey, most of which are young clusters in the disk of M31; eight others are located in distant halos more than 25 kiloparsec (1 kiloparsec≈ 32,62 light-year) from the center of M31, and they are old globular clusters.