马氏距离度量LAMOST早型星光谱的分类研究

随着天文大数据不断积累，我国大天区多目标光纤光谱望远镜LAMOST已完成6年的大规模巡天观测，获得DR5数据集已达到900多万条光谱，其中含有观测比例较低的早型恒星光谱，具备重要的研究价值。利用准确的恒星分类模板库可提升恒星的分类精度与可靠性，由于LAMOST第一年的巡天光谱中并没有完整覆盖B型恒星包含的所有子类型，造成后续观测数据分类的子类型范围受限。依据LAMOST已发布DR5数据中B型恒星光谱为研究对象，选取ELODIE发布的B型恒星实测光谱模板库来检测LAMOST在用的分类光谱。首先完成ELODIE发布37条B型光谱模板的相关性分析，去掉相关性弱的三条光谱后，筛选出ELODIE 34条B型恒星实测模板作为中心，通过计算LAMOST DR5发布的绝大多数被标记为B6型（7 662条）和B9型（3 969条）实测光谱的马氏距离，经有监督聚类LAMOST早型恒星光谱数据，标记13个子类型在涵盖B2-B9子类的34条ELODIE光谱模板中的分布。经线性分析判别每条谱线子类型的类内距离，确保波长覆盖范围和分辨率与LAMOST数据完全一致，去掉距离数值偏差较大的数据，计算相应子类的平均谱线，得到LAMOST源于DR5观测数据早型B型恒星的13条子类型光谱分类模板，为后期完善模板提供较好的参考性。

Study on the Classification of LAMOST Early Stellar Spectrum Template by Mahalanobis Distance

With the continuous accumulation of astronomical data, Large Sky Area Multi-Object Fiber Spectroscopy Telescope (LAMOST) has completed six years of large-scale sky surveys. The DR5 dataset has obtained more than 9 million spectrum data, including early-type stellar spectra with the lower observation proportion. The correct stellar classification template library can improve the classification accuracy. The currently used classification templates in LAMOST pipeline which don’t completely cover all the subtypes such as B-type stars, because they were constructed using DR1 data without enough early type stars. In this paper, the B-type stellar spectra in LAMOST DR5 have been collected as our research object, and the reference B-type spectra are from the library of ELODIE. Firstly, we complete the correlation analysis of 37 spectra of ELODIE B-type stars. After removing three weakly correlated spectra, 34 spectra of ELODIE B-type stars spectra were selected as the cluster center. The majority of the published LAMOST DR5 labels were marked as B6 (7662) and B9 (3969) and spectra were measured by Mahalanobis distances, with the supervised clustering, 34 LAMOST early-type stellar spectral data were marked as 13 subtypes according to ELODIE labels covering from B2 to B9 subclasses. The intra-class distance of each spectral subtype is determined by linear analysis to ensure that the wavelength coverage and resolution are completely consistent with the LAMOST data. The average spectral line of the corresponding subclass is calculated removing the outliers, thus 13 subtype spectral classification templates of B-type stars provide a good reference for later template completion.