Classification of Galaxy Morphology Based on Multi-Channel Deep Residual Networks

Abstract

In the process of studying the aggregation and evolution of cosmic galaxies, galaxy morphology is an important parameter to be considered. With the rapid development of deep learning and artificial intelligence technology, galaxy morphology classification based on deep learning framework convolutional neural network emerges. In this paper, three typical deep learning frameworks, AlexNet, ResNet50 and VGGNet-E, are used to classify the galaxy images of Galaxy Zoo 2. The performance of the three deep learning frameworks is evaluated by accuracy and loss rate, and it is found that ResNet50 works best. In this paper, a multi-channel depth residual network framework ResNet-Core is designed based on the ResNet50 network structure. This structure can deepen the extraction of detailed features through the control of convolution kernel. The experimental results show that compared with AlexNet,ResNet50,VGGNet-E deep learning neural network models,ResNet-Core model has better classification performance and better robustness.