The Galaxy Morphology Classification Based On Deep Learning

Galaxies are orbiting systems composed of numerous stellar systems and interstellar dust.Most galaxies have more obvious geometric features,and different galaxy morphologies correspond to different physical properties.Through the evolution of galaxy structure over time,we can understand the formation and evolution process of galaxies,in which how to correctly classify galaxies and understand the morphological structure of galaxies is very important to study the physical properties of galaxies.Deep learning methods can realize automatic feature extraction and identification,which greatly accelerate the efficiency of data processing and analysis.Therefore,this paper uses deep learning methods to classify galaxy morphology.Capsule networks have spatial feature extraction capability,which can fully exploit the morphological features of galaxies,and have achieved some good results in galaxy classification tasks,but the basic capsule networks still have room for improvement,such as high requirements for data distribution,poor generalization ability,and difficult training.In this paper,based on the traditional capsule network,we use a multi-scale parallel convolutional layer to sample galaxy images with different granularity and extract multi-scale features of galaxy morphology,so that the network can learn more comprehensive feature information and improve the feature learning and representation capability of the network.In addition,this paper uses a Sigmoid Routing to replace the traditional routing algorithm to improve the classification performance of the model.Experimental analysis is performed for galaxy image data from galaxy zoo,and the results show that the final classification accuracy of our model for galaxy morphology is 97.01%,Recall is 98.16%,and F1-score is 96.39%,compared with other methods such as Deleman,Res Net-26,and NODE,our method is more effective for galaxy classification and performance better.Vision Transformer(Vi T)has gained more and more attention from researchers because of its powerful sequence modeling capability and global information perception.In this paper,we use a multi-scale feature fusion Vision Transformer galaxy morphology classification model for the morphological classification of galaxies.In this work,a lightweight attention mechanism is used to fuse features of different scales for the simple galaxy morphological features and monotonous background,and the galaxy data is classified into 7 classes to explore the galaxy classification task under more classes.The results show that the final classification accuracy of the model is93.03%,Recall is 92.56%,and F1-score is 94.69%.Compared with the base Vi T,the model accuracy is improved by nearly 10 percentage points.The visual Transformer galaxy morphology classification model with multi-scale feature fusion achieves good classification results in the classification task of seven types of galaxies,which can be helpful for more complex galaxy classification tasks in the future.This paper also explores the visualization of features in the learning process of deep learning models,using high-dimensional data downscaling techniques to downscale and visualize high-dimensional features,trying to extract the phenomena and laws embedded in galaxy morphology data from high-dimensional data,and quickly grasp the nature of the physical external representation of galaxy morphology data,providing more useful technical and theoretical support for future galaxy morphology research.In this paper,two improved deep learning algorithms are used to classify galaxy morphology data from galaxy zoos,and good classification results are achieved,providing a new approach to the processing and analysis of galaxy observation data.In addition,this paper uses the t-SNE algorithm to visualize and analyze the features of high-dimensional data,trying to explore the physical representation of highdimensional data,to explore the physical laws behind the high-dimensional data,and to provide more powerful technical support for the study and analysis of galaxies.