A new method for clustering of boundary spectra

摘要

The stellar spectral data taken by LAMOST (Large Sky Area Multi-Object Fiber Spectroscopic Telescope) include multiple types, some of which that fall between two spectral classes, namely boundary spectra. Due to the massive and high dimensional nature of spectra data, it will take a lot of time and energyto cluster these spectra by manual operation alone. To address this problem, a new clustering method based on influence space is presented in this paper. First, we introduce the concept of influence space to reduce the amount of data involved in the operation, and reduce the dimension of the data by extracting the main feature lines. Second, a novel method for initial cluster center selection is applied. Next, based on the selected initial cluster centres, other spectra are clustered by running K-means algorithm on the whole data set. The experimental results indicate that the initial cluster centres obtained by this method are of higher quality and the problem of boundary spectra clustering is also well solved.