Galaxy morphology classification based on ResNeXt

Abstract

The morphology of galaxies can reflect the physical properties of galaxies themselves, and the classification of their morphology plays an important role in the subsequent analysis and research.In this paper, we use the photometry image of galaxy in GalaxyZoo2, select the data set according to the threshold and perform data augmentation, and apply ResNeXt to the classification of galaxy morphology, which realizes the automatic extraction, recognition and classification of galaxy morphological features.Based on the results of ResNeXt's galaxy morphology classification, five groups of comparative experiments are carried out.The five groups of comparison experiments include comparing different versions of ResNeXt model, comparing classical convolutional neural network model, comparing the latest image classification model in the last two years, comparing the simplest convolutional neural network model, and comparing the human eye.The experimental results show that the galaxy morphology classification accuracy based on ResNeXt101 network model is the highest.