The Impact of Batch Size on Deep Learning-Based Galaxy Morphology Classification Using ResNet50

Abstract

Galaxy morphology classification is a fundamental task in astronomy, which helps to gain a deeper understanding of the formation and evolution of galaxies. This paper adopts a convolutional neural network as the backbone model to achieve the basic requirements of galaxy classification. In this study, a frozen convolutional base ResNet50 is used as the backbone network to systematically explore the impact of learning rate and batch size on the classification accuracy of 10 types of galaxy morphologies. The results show that when the batch sizes are 16, 32, 64, and 128, the overall accuracies are 74.75%, 74.93%, 75.01%, and 74.90% respectively. A medium batch size of 64 achieves the best balance between optimization stability and generalization ability. This study provides a reference direction for the impact of batch size on the model and offers new ideas for better solving the problem of galaxy classification. Future research may extend this work by incorporating transfer learning or self-supervised methods to further enhance model robustness and accuracy across larger astronomical datasets.