光谱与测光数据融合算法在变星分类上的应用

Application of Spectral and Metering Data Fusion Algorithm in Variable Star Classification

天文学上把亮度随时间变化的恒星称为变星。它对于研究星系的距离,恒星的演化以及恒星在不同阶段的性质具有非常重要的意义。目前对变星的识别主要依靠长时间观测其亮度变化,并结合对恒星的光谱进行分析才能最终完成认证。这项工作需要天文学家投入大量时间,难以开展大规模分类。针对上述问题本文提出了一种将测光图像与一维光谱进行数据融合用于对变星进行分类的方法——光谱-测光融合网络(ASPF-Net)。该网络由C1网络和C2网络两部分组成,其中C1是用于提取光谱特征的一维卷积神经网络,C2是用于提取测光数据特征的二维卷积神经网络;最后将两者提取到的特征进行融合,用一个全连接前馈神经网络完成分类。该研究在对食变星、脉冲变星和标准星分类问题上进行了实验。实验数据均来自于斯隆数字巡天项目(SDSS),该项目包含了测光图像和光谱两种数据。对于光谱数据本文选取波长在380.0~680.0 nm范围内的流量值。测光图像由:u、g、r、i和z共5个波段数据组成,对应的中心波长分别为:355.1、468.6、616.6、748.0和893.2 nm。相比于传统的利用其中三个波段合成RGB图像,原始SDSS数据拥有更高的灰度等级。为了方便网络训练,对测光数据和光谱数据均做了标准化处理。分类性能分析方面,使用了精确率,召回率,F1值和平均准确率四个指标进行评估。提出的光谱-测光融合网络(ASPF-Net)在针对食双星、脉冲变星和标准星的分类任务,精确率分别为:91.1%、92.8%和98.2%。实验证明,数据融合之后的分类性能优于单独使用光谱数据或测光数据的分类性能。说明将光谱数据和测光数据结合起来对变星进行分类是一种有效的方法,这为今后的变星的分类提供了一种新的思路和方法。

In astronomy,stars whose brightness changes with time are called variable stars.It is of great significance to study the distance of galaxies,the evolution of stars and the properties of stars in different stages.At present,the identification of variable stars mainly depends on observing their brightness changes for a long time and analyzing the spectrum of stars.This work requires astronomers to invest much time,so it is not easy to carry out large-scale classification.A data fusion method of photometric image and one-dimensional spectrum is proposed to classify variable stars.Experiments are carried out on identifying eclipsing variable stars,pulsar variable stars and standard stars.The Sloan Digital Sky Survey project includes photometric images and spectral data.For spectral data,this paper selects the flow value in the wavelength range of 380.0~680.0 nm.The photometric image comprises five data bands:u,g,r,i and z.The corresponding center wavelengths are 355.1,468.6,616.6,748.0 and 893.2 nm respectively.The photometric value is generally distributed between 0 and 200.For the image part,this paper uses the data of five bands for classification,and locates the position of the target star in the photometric image through the star catalogue.In order to facilitate network training,photometric and spectral data are standardized.The method of combining photometric image and spectral data for classification proposed in this paper uses four indicators:accuracy,recall,F1 value and average accuracy.The experimental results after data fusion are better than those using spectral or photometric data alone.For the classification tasks of eclipsing binaries,pulsars and standard stars,the accuracy rates are 91.1%,92.8%and 98.2%respectively.Experiments show that the combination of image data and photometric data is an effective method for star classification,which provides a new idea and method for star classification in the future.