Identification of Extragalactic Globular Clusters Using Machine Learning Techniques

Abstract

Classification of extragalactic globular clusters (GC) is a process that requires a considerable amount of time from experts. To facilitate this laborious process, in this study, we studied a machine learning-based classification pipeline and demonstrated its application on the classification of GC and other celestial object classes in the galaxy M81. Due to the lack of annotated data, we also need to infer pseudo labels of celestial objects in the images based on data clusters. The proposed pipeline starts with a feature extraction step using the autoencoder method. Then, after obtaining the feature vector, the objects are grouped into 12 clusters by a clustering algorithm. After that, the cluster labels are used as pseudo labels for training a classification model. The experimental results based on 10-fold cross-validation showed that the combination of subspace clustering and deep neural network reached the best average accuracy of 78.68% among the three clustering algorithms and the five classification models under comparison.