Classification of celestial spectral based on improved density clustering

With the rapid development of large-scale survey plans, the automatic classification of massive celestial spectrum data has urgently become an important research project. This paper presents the principle that based on density and improved density clustering algorithm - The Manhattan Distance density algorithm(MD-DBSCAN), applied to a variety of spectral data. According to the United States SDSS-DR8 astronomical spectral data, compared to three algorithms of DBSCAN, NED-DBSCAN, MD-DBSCAN performance, then give the corresponding conclusions, the average classified stable number of the MD-DBSCAN algorithm is smaller and computing time is shorter.