Searching for Hot Subdwarf Stars from the LAMOST Spectra. III. Classification of Hot Subdwarf Stars in the Fourth Data Release of LAMOST Using a Deep Learning Method
Searching for Hot Subdwarf Stars from the LAMOST Spectra. III. Classification of Hot Subdwarf Stars in the Fourth Data Release of LAMOST Using a Deep Learning Method
复制标题
DOI：
10.3847/1538-4357/ab4c47
复制
发表时间：
2018-05
期刊：
影响因子：
--
通讯作者：
Yude Bu;Jingjing Zeng;Z. Lei;Z. Yi
中科院分区：
文献类型：
--
作者：
Yude Bu;Jingjing Zeng;Z. Lei;Z. Yi
热亚矮星是位于水平分支蓝色一端的核心氦燃烧恒星，也被称为极水平分支。热亚矮星的光谱可以提供关于恒星大气参数的详细信息，如有效温度、重力和氦丰度，这有助于阐明热亚矮星的天体物理和统计性质。这些性质为恒星的理论模型提供了重要的约束。从大天空面积多目标光纤光谱望远镜(LAMOST)获得的光谱数据中识别热亚矮星，可以显著增加样本规模，帮助我们更好地了解热亚矮星的性质。在这项研究中，我们提出了一种利用卷积神经网络和支持向量机(CNN+SVM)从LAMOST光谱中选择热亚矮星的新方法。将CNN+支持向量机应用于LAMOST数据版本4中的样本数据，得到了76.98%的F1得分。与线性判别分析、k近邻等机器学习算法的比较表明，基于CNN+支持向量机的方法取得了较好的效果。因此，它是一种非常适合于在大型光谱调查中寻找热亚矮星的方法。最后，我们对如何确定我们提出的方法的最优超参数进行了广泛的讨论。
Hot subdwarf stars are core He burning stars located at the blue end of the horizontal branch, which is also known as the extreme horizontal branch. The spectra of hot subdwarf stars can provide detailed information on stellar atmospheric parameters, such as the effective temperature, gravity, and abundances of helium, which can help clarify the astrophysical and statistical properties of hot subdwarf stars. These properties provide important constraints on the theoretical models of stars. The identification of hot subdwarf stars from the spectral data obtained by the Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) can significantly increase the sample size and help us to better understand the nature of hot subdwarf stars. In this study, we propose a new method to select hot subdwarf stars from LAMOST spectra using convolutional neural networks and a support vector machine (CNN+SVM). By applying CNN+SVM to sample data selected from LAMOST Data Release 4 we obtain an F1 score of 76.98%. A comparison with other machine-learning algorithms, such as linear discriminant analysis and k-nearest neighbors, demonstrates that an approach based on CNN+SVM obtains better results than the others. Therefore it is a method well suited to the problem of searching for hot subdwarf stars in large spectroscopic surveys. Finally, we include an extensive discussion on how we determined the optimal hyperparameters of our proposed method.