利用暗能量巡天将自动化形态分类推向极限
星系天体物理
2021-03-17 v2 宇宙学与河外天体物理
摘要
我们使用监督深度学习算法给出了暗能量巡天(DES)数据发布1(DR1)中约2700万星系的形态分类。分类方案区分：(a)早型星系(ETGs)与晚型星系(LTGs)；(b)正面星系与侧向星系。我们的卷积神经网络(CNNs)在具有已知分类的小部分DES天体上训练。这些天体通常有；我们通过模拟具有确定分类的较亮天体在高红移下的形态，将较暗天体建模至 mag。CNNs在其训练集上达到97%的准确率(至)，表明它们能够比人眼更准确地恢复特征。我们随后用训练好的CNNs分类绝大多数其他DES图像。最终星表为每个分类方案包含五个独立的CNN预测，有助于判断CNN预测是否稳健。我们对星表中约87%和73%的对象分别获得了ETG vs. LTG与侧向 vs. 正面模型的可靠分类。结合两种分类(a)与(b)有助于提高ETG样本纯度并识别侧向透镜状星系(作为高椭率的ETGs)。在可比较处，我们的分类与Sérsic指数(\textit{n})、椭率()和光谱型相关性极好，即便对较暗星系亦然。这是迄今最大的自动化星系形态多波段星表。
引用
@article{arxiv.2012.07858,
title = {Pushing automated morphological classifications to their limits with the Dark Energy Survey},
author = {J. Vega-Ferrero and H. Domínguez Sánchez and M. Bernardi and M. Huertas-Company and R. Morgan and B. Margalef and M. Aguena and S. Allam and J. Annis and S. Avila and D. Bacon and E. Bertin and D. Brooks and A. Carnero Rosell and M. Carrasco Kind and J. Carretero and A. Choi and C. Conselice and M. Costanzi and L. N. da Costa and M. E. S. Pereira and J. De Vicente and S. Desai and I. Ferrero and P. Fosalba and J. Frieman and J. García-Bellido and D. Gruen and R. A. Gruendl and J. Gschwend and G. Gutierrez and W. G. Hartley and S. R. Hinton and D. L. Hollowood and K. Honscheid and B. Hoyle and M. Jarvis and A. G. Kim and K. Kuehn and N. Kuropatkin and M. Lima and M. A. G. Maia and F. Menanteau and R. Miquel and R. L. C. Ogando and A. Palmese and F. Paz-Chinchón and A. A. Plazas and A. K. Romer and E. Sanchez and V. Scarpine and M. Schubnell and S. Serrano and I. Sevilla-Noarbe and M. Smith and E. Suchyta and M. E. C. Swanson and G. Tarle and F. Tarsitano and C. To and D. L. Tucker and T. N. Varga and R. D. Wilkinson},
journal= {arXiv preprint arXiv:2012.07858},
year = {2021}
}
备注
Accepted for publication in MNRAS (2021 February 22); 17 pages, 16 figures