通过对恒星光谱进行分析可以研究银河系的演化与结构等科学问题, 光谱分类是恒星光谱分析的基本任务之一。 提出了一种结合非参数回归与Adaboost对恒星光谱进行MK分类的方法, 将恒星按光谱型和光度型进行分类, 并识别其光谱型的次型。 恒星光谱的光谱型及其次型代表了恒星的表面有效温度, 而光度型则代表了恒星的发光强度。 在同一种光谱型下, 光度型反映了谱线形状细节的变化, 因此光度型的分类必须在光谱型分类基础上进行。 本文把光谱型的分类问题转化为对类别的回归问题, 采用非参数回归方法进行恒星光谱型和光谱次型的分类; 基于Adaboost方法组合一组K近邻分类器进行光度型分类, Adaboost将一组弱分类器加权组合产生一个强分类器, 提升光度型的识别率。 实验验证了所提出分类方法的有效性, 光谱次型识别的精度达到0.22, 光度型的分类正确率达到84%以上。 实验还对比了两种KNN方法与Adaboost方法的光度型分类, 结果表明, 利用KNN方法对光度型分类精度低, 而基于弱分类器KNN的Adaboost方法将识别率大幅提升。

With the analysis of stellar spectra, the evolution and structure of the Milky Way galaxy is studied. Spectral classification is one of the basic tasks of stellar spectral analysis. In this paper, a method of MK classification based on non parametric regression and Adaboost for stellar spectra is proposed, and the stars are classified according to the luminosity type, spectral type as well as the spectral subtype. The spectral type of the stellar spectrum and its sub type represent the effective temperature of the star, while the luminosity type represents the luminous intensity of the star. In the same spectral type, the luminosity type reflects the variation of the shape details of the spectral line, so the classification of the photometric type must be based on the spectral type classification. The spectral type classification is transformed as a regression problem of class label, and the type and subtype of the stellar spectra are recognized with non parametric regression method. The luminosity type of the stellar spectra is recognized using Adaboost method which combines a group of K nearest neighbor classifiers. Adaboost generates a strong classifier with weighted combination of a group of weak classifiers to improve the recognition rate of the luminosity type. Experimental results validate the proposed method. The accuracy of spectral subtype recognition is up to 0.22, and the correct rate of the luminosity type classification is 84% above. Two KNN methods are compared with Adaboost method on luminosity recognition. The results show that the recognition rate can be greatly enhanced with the Adaboost method and using KNN.