通过无监督-监督混合方法揭示星系形态
Unveiling Galaxy Morphology through an Unsupervised-Supervised Hybrid Approach
AI总结:
本文提出一种融合无监督聚类与监督卷积神经网络的星系形态分类方法,并引入EGG形态指标,在SDSS数据上实现95%无监督分类准确率。
摘要:
星系形态为星系的演化路径和基础物理提供了重要见解。随着Euclid和Vera C. Rubin等巡天项目使天文数据不断增长,需要工具来分类和分析将要观测到的海量星系。在这项工作中,我们引入了一种新的分类技术,将基于形态学指标的无监督聚类与监督卷积神经网络的可扩展性相结合。我们深入比较分析了著名的CAS(集中度、不对称性和平滑度)指标与我们新提出的EGG(熵、基尼系数和梯度模式分析)指标。我们选择EGG系统的原因在于其面向分离的指标,能够最大化形态类别对比度。我们利用指标与形态类别之间的关系,实现了无监督聚类与监督分类之间的内部一致性。将我们的方法应用于斯隆数字巡天数据,我们获得了纯无监督分类95%的总体准确率,并且在复现T-Type和目视分类星系目录时分别获得了88%和89%的准确率,说明了该方法的实用性。此外,对哈勃空间望远镜数据的应用预示着对更高红移范围进行无监督探索的潜力。一项显著成就是我们在无监督分类中达到95%的准确率,这一结果在与传统机器学习对比时具有竞争力,在与深度学习基准相比时也紧随其后。
英文摘要:
Galaxy morphology offers significant insights into the evolutionary pathways and underlying physics of galaxies. As astronomical data grows with surveys such as Euclid and Vera C. Rubin , there is a need for tools to classify and analyze the vast numbers of galaxies that will be observed. In this work, we introduce a novel classification technique blending unsupervised clustering based on morphological metrics with the scalability of supervised Convolutional Neural Networks. We delve into a comparative analysis between the well-known CAS (Concentration, Asymmetry, and Smoothness) metrics and our newly proposed EGG (Entropy, Gini, and Gradient Pattern Analysis). Our choice of the EGG system stems from its separation-oriented metrics, maximizing morphological class contrast. We leverage relationships between metrics and morphological classes, leading to an internal agreement between unsupervised clustering and supervised classification. Applying our methodology to the Sloan Digital Sky Survey data, we obtain 95% of Overall Accuracy of purely unsupervised classification and when we replicate T-Type and visually classified galaxy catalogs with accuracy of 88% and 89% respectively, illustrating the method's practicality. Furthermore, the application to Hubble Space Telescope data heralds the potential for unsupervised exploration of a higher redshift range. A notable achievement is our 95% accuracy in unsupervised classification, a result that rivals when juxtaposed with Traditional Machine Learning and closely trails when compared to Deep Learning benchmarks.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。