DB white dwarf template construction for LAMOST 1D pipeline

We adopt a machine learning (ML) method, introduced in detail in our previous work, to mine spectroscopicallyconfirmed DB white dwarf (DBWD) from LAMOST Data Release(DR) 5. The unique features of DBWD are extracted from between known DB spectra and all other released data. We take advantage of these DBWD samples and features by classifying a certain amount of LAMOST spectral data by LAMOST 1D Pipeline. At first, two groups of clustering centers are produced as DBWD templates using k-means. Then, we build four control groups, whether to consider feature location and which clustering centers are employed, to conduct classification tests. The experiment demonstrates that taking particular features as weights of spectral data could improve classification accuracy.