基于深度信念网络的天体光谱自动分类研究

把深度信念网络应用于天体光谱的分类.首先,使用小波变换对光谱数据进行降噪预处理,其次,采用PCA对光谱数据进行特征值提取降维,然后建立深度信念网络模型并构造分类器,最后使用该分类器对美国斯隆巡天项目的天体光谱数据进行激变变星的分类研究,并与受限波尔兹曼机网络进行了对比研究.由于深度信念网络对数据有深层次的学习能力,采用深度信念网络对天体光谱进行分类有一定优势.实验结果证明了分类方法的有效性.

Automatic classification of star spectra based on the deep belief network

This paper applies the deep belief network to the classification of astronomical spectra. First of all,the wavelet transform is used for the preliminary denoising of the spectral data. Then,the Principal Component Analysis( PCA) is used for the dimensionality reduction of the feature-value acquisition of the spectral data. Finally,this classifier is used for the study of some Cataclysmic Variable Stars in the Sloan Digital Sky Survey and then gives it a comparative study with the Restricted Boltzmann Machines( RBM). Because the deep belief network has data-based deep learning skills,it has the advantage of classifying the astronomical spectra,which has been proved in this study.