恒星的分类对了解恒星和星系形成与演化历史具有重要的研究价值。 面对大型巡天计划及由此产生的海量数据, 如何迅速准确地将天体自动分类显得尤为重要。 通过对SDSS DR9的恒星光谱数据进行深度置信神经网络(DBN)、 神经网络和支持向量机(SVM)等算法分类的对比, 分析三种自动光谱分类方法在恒星分类上的适用性。 首先利用上述三种方法对K, F恒星进行识别分类, 然后再分别对K1, K3和K5次型和F2, F5, F9次型识别, 最后基于SVM支持向量机的二次分类模型, 利用Ｋ次型的数据, 构建剔除不属于Ｋ次型的模型。 结果表明: 深度置信网络对K, F型恒星分类效果较好, 但是对K, F次型的分类效果不佳; SVM支持向量机在K, F型恒星分类以及相应的次型分类都具有较好的识别率, 对K, F型分类效果要好于K, F次型的分类效果; BP神经网络对K, F型恒星以及其次型的识别一般; 在剔除不属于Ｋ次型实验中, 剔除率高达100%, 可知SVM能够对未知的光谱数据进行筛选与分类。

The stellar classification was an important research field for understanding the formation and evolution of stars and galaxies. With large sky surveys and its massive data, the speed and accuracy of the celestial automatic classification was very important. The depth confidence neural network (DBN), support vector machines (SVM) and BP neural networks used in the star classification were compared in this paper. And the applicability of star classification with these three methods was analyzed. First, K, F stars are classified according to the depth of confidence neural network and BP neural network and support vector machine.Then the K1, K3, K5 sub-type and F2, F5, F9 sub-type were separately identified. Finally, the data which did not belong to the k sub-type were excluded by a secondary classification model based on SVM support vector machine . The results shows that: the depth of belief networks is better for K, F-type star classification, but it is poor for K, F sub-type classification results; The recognition rate of SVM is high for the K, F-type stars and the classification effects of this method is better for K, F-type stars than the corresponding sub-type stars by comparison; The recognition rate of BP neural network is ordinary general for K, F-type stars and their sub-types. The experiment showed that the accuracy of excluding non-k-sub-type data can be up to 100% which indicates that the unknown spectral data can be screened and classified with SVM.