GReFC-Net: an automated method for measuring structural features of spiral galaxies

Abstract

The spiral structure is an important morphology within galaxies, providing information on the formation, evolution, and environment of spiral galaxies. The number of spiral arms is one of the important parameters to describe the morphology of spiral galaxies. In this project, we study the classification of spiral galaxies by the number of spiral arms based on deep learning algorithms. The data set for this project consists of eligible samples from Galaxy Zoo 2 and Galaxy Zoo DECaLS. To better identify the texture features of the spiral arms, we designed a convolutional neural network model incorporating Gabor filter (Gabor Residual Filtering Convolutional Net, GReFC-Net), and used other networks for 3 and 4-way classifications. In the 3-way case, the GReFC-Net algorithm achieves the highest precision, recall, F1-score, and AUC value, which are 96.25%, 96.23%, 96.21%, and 0.9937. In the 4-way case, the GReFC-Net algorithm has the highest recall, F1-score and AUC value, which are 95.57%, 95.42% and 0.9957. The interpretability of GReFC-Net is analyzed by the SHAP method, and the results show that the network can identify the spiral arm structure of spiral galaxies well. It can be seen that the GReFC-Net algorithm can be effectively applied to the automatic measurement task of spiral arm structure in a large number of spiral galaxies.