使用SOFM方法进行恒星光谱自动分类

ＳＯＦＭ是人工神经网络的非监督算法,可以将数据组织到一个特征图上,而保存大多数原始数据空间的拓扑特征,使用这种方法进行恒星光谱自动分类,分类结果与哈佛序列十分相似,ＳＯＦＭ方法应该是进行大数量恒星光谱样本在线分类的有用方法,它能够自动执行,因此可用于处理大数量天体光谱。     

晚型星计算机自动光谱分类

在对恒星形成区（ＳＦＲ）中ＲＯＳＡＴ选弱发射线Ｔ Ｔａｕｒｉ星（ＷＴＴＳ）进行光谱证认的过程中,发展了一套基于北京天台２．１６米光学望远镜及其ＯＭＲ卡焦光谱仪系统的晚型星中色散（５０Ａ／ｍｍ）光谱计算机自动光谱分类方法。对ＲＯＳＡＴ选ＷＴＴＳ候选体中晚型星进行自动光谱分类的结果表明,一般情况下,光谱型的分类精度可达±１个次级,个别源为±２个次级。该自动光谱分类方法同时适用于其它光谱晚型星。     

晚型星计算机自动光谱分类

在对恒星形成区（ＳＦＲ）中ＲＯＳＡＴ选弱发射线ＴＴａｕｒｉ星（ＷＴＴＳ）进行光谱证认的过程中,发展了一套基于北京天文台２．１６米光学望远镜及其ＯＭＲ卡焦光谱仪系统的晚型星中色散（５０Ａ／ｍｍ）光谱计算机自动光谱分类方法．对ＲＯＳＡＴ选ＷＴＴＳ候选作中晚型星进行自动光谱分类的结果表明,一般情况下,光谱型的分类精度可达±１个次级,个别源为±２个次级．该自动光谱分类方法同时适用于其他光谱晚型星．     

基于样条函数的恒星光谱自动归一化方法

恒星的观测谱一般由连续谱、谱线和噪声组成,其中连续谱是黑体辐射导致的辐射流量随波长变化的光滑连续光谱。光谱分类及恒星物理参数估计等研究依赖于连续谱及谱线信息的准确提取。因此光谱数据处理的工作主要是拟合连续谱,并通过对光谱进行归一化来提取谱线特征。连续谱拟合的方法主要有多项式拟合、中值滤波、小波滤波等。已有的方法在低信噪比、宇宙线信号干扰、存在发射线等情况下,有不同程度的局限性,体现在鲁棒性和准确度上。目前,针对郭守敬望远镜的10 7条光谱没有自动化方法应用到归一化上的问题,研究并开发一种适用于不同的温度、信噪比及波长覆盖范围,并能够自动化处理的恒星光谱归一化方法,显得十分迫切。在仔细分析不同类型光谱的基础上,提出了一种基于固定窗口划分的连续谱拟合方法。该方法对光谱中能够体现连续谱特征的数据点进行筛选提取,通过细微地控制样条函数平滑度产生更加准确的连续谱。使用郭守敬望远镜中不同光谱型、温度范围、波长覆盖范围的光谱进行实验,结果表明,该方法具有良好的精度和普适性。     

共生星冷星光谱分类

在近工外波段,对１０颗共生星,２７颗Ｋ－Ｍ型比较星进行了分辨本领对晚于Ｍ２的晚型星,出现了ＴｉＯ三重头吸收带,发现吸收带深度对温度很敏感而对重力不敏感,用最小二乘法拟合了光谱型对ＴｉＯ吸收带深度的曲一,拟合曲线的标准方差为０．２２光谱分型,并由此给出６颗含晚于Ｍ２巨星的共生星光谱型。另外还测量了该波段上的ＣａⅡＤＧ ＴＧＪ 线,ＦｅⅠ,ＴｉⅠ,ＮａⅠ等众谱线的等值宽度。发现ＣａⅡ三重线对光度级很咸     

基于卷积神经网络的恒星光谱型和光度型的分类模型

恒星光谱分类是天文学中一个重要的研究问题.对于已经采集到的海量高维恒星光谱数据的分类,采用模式匹配方法对光谱型分类较为成功,但其缺点在于标准恒星模版之间的差异性在匹配实际观测数据中不能体现出来,尤其是当需要进行光谱型和光度型的二元分类时模版匹配法往往会失败.而采用谱线特征测量的光度型分类强烈地依赖谱线拟合的准确性.为了解决二元分类的问题,介绍了一种基于卷积神经网络的恒星光谱型和光度型分类模型(Classification model of Stellar Spectral type and Luminosity type based on Convolution Neural Network, CSSL CNN).这一模型使用卷积神经网络来提取光谱的特征,通过注意力模块学习到了重要的光谱特征,借助池化操作降低了光谱的维度并压缩了模型参数的数量,使用全连接层来学习特征并对恒星光谱进行分类.实验中使用了大天区面积多目标光纤光谱天文望远镜(Large Sky Area Multi-Object Fiber Spectroscopy Telescope, LAMOST)公开数据集Data Release 5 (DR5,用了其中71282条恒星光谱数据,每条光谱包含了3000多维的特征)对该模型的性能进行验证与评估.实验结果表明,基于卷积神经网络的模型在恒星的光谱型分类上准确率达到92.04%,而基于深度神经网络的模型(Celestial bodies Spectral Classification Model, CSC Model)只有87.54%的准确率; CSSL CNN在恒星的光谱型和光度型二元分类上准确率达到83.91%,而模式匹配方法MKCLASS仅有38.38%的准确率且效率较低.     

基于胶囊网络的恒星光谱分类研究

大型巡天项目的快速发展,产生大量的恒星光谱数据,也使得实现恒星光谱数据的自动分类成为一项具有挑战性的工作.提出一种新的基于胶囊网络的恒星光谱分类方法,首先利用1维卷积网络和短时傅里叶变换将来源于LAMOST(Large Sky Area Multi-Object Fiber Spectroscopy Telescope)Data Release 5(DR5)的F5、G5、K5型1维恒星光谱转化成2维傅里叶谱图像,再通过胶囊网络对2维谱图像进行自动分类.由于胶囊网络具有保留图像中实体之间的分层位姿关系和无需池化层的优点,实验结果表明:胶囊网络具有较好的分类性能,对于F5、G5、K5型恒星光谱的分类,准确率优于其他分类方法.     

The Cannon:一种基于光谱数据的恒星参数测量方法

新一代大规模光谱巡天项目产生了近千万条低分辨率恒星光谱,基于这些光谱数据,介绍一种名为The Cannon的机器学习方法。该方法完全基于已知恒星大气参数(有效温度、表面重力加速度和金属丰度等)的光谱数据,通过数据驱动来构建特征向量,建立光谱流量特征和恒星参数的函数对应关系,进而应用到观测光谱数据中,实现对恒星光谱的大气参数求解。The Cannon的主要优势为不直接基于任何恒星物理模型,适用性更广;由于使用了全谱信息,即便对于低信噪比光谱也能得到较高可信度的参数结果,该算法在大规模恒星光谱的数据处理和参数求解方面具有明显的优势。此外,还利用The Cannon得到LAMOST光谱数据中K巨星和M巨星的恒星参数。     

Automatic Determination of Stellar Atmospheric Parameters Using Neural Networks and Instance-Based Machine Learning

In this article we show how machine learning methods can beeffectively applied to the problem of automatically predictingstellar atmospheric parameters from spectral information, a veryimportant problem in stellar astronomy. We apply feedforwardneural networks, Kohonen's self-organizing maps andlocally-weighted regression to predict the stellar atmosphericparameters effective temperature, surface gravity and metallicityfrom spectral indices. Our experimental results show that thethree methods are capable of predicting the parameters with verygood accuracy. Locally weighted regression gives slightly betterresults than the other methods using the original dataset asinput, while self-organizing maps outperform the other methods when significant amounts of noise are added. We also implemented a heterogeneous ensemble of predictors, combining the results given by the three algorithms. This ensemble yields better results than any of the three algorithms alone, using both the original and the noisy data.     

用Monte Carlo方法拟合超新星SN 1993J的早期光谱

通过把从光球辐射出来的连续谱能量分成大量的光子包，然后用MonteCarlo方法跟踪这些光子包从光球向超新星包层外运动的过程，这一方法能很好地解决超新星的光谱拟合中的谱线覆盖问题，由于同时编制了用于构造超新星包层中的密度结构和丰度分布的程序，以及大量的用于比较观测和理论拟合的绘图和支持程序，使我们的MonteCarlo光谱拟合软件包能够很好地运用于超新星的早期光谱研究。本文给出了SN1993J在1993年4月13日的观测光谱的运行实例。     

Automatic Normalization and Equivalent-Width Measurement of High-Resolution Stellar Spectra

It is well known that normalization, radial velocity correction and equivalent-width measurement of high-resolution stellar spectra are time-consuming work. In order to improve the efficiency we present an automatic method for these routines. The continuum is determined by fitting the 'high points' in the spectrum. After continuum normalization, the program automatically searches for the position of the Ha line and obtains a rough radial velocity, then computes an accurate radial velocity by cross-correlation between the given spectrum and the solar spectrum. In this method, the equivalent-width is automatically measured using Gaussian fitting. A comparison between our results and those from traditional analysis shows that the typical error for equivalent width is around 3.8% in our method. Developing such automatic routines does not mean to replace the interactive reduction method: it is just for a quick extraction of information from the spectra, especially those obtained in large sky surveys.     

Automated Stellar Classification for Large Surveys with EKF and RBF Neural Networks

An automated classification technique for large size stellar surveys is proposed. It uses the extended Kalman filter as a feature selector and pre-classifier of the data, and the radial basis function neural networks for the classification. Experiments with real data have shown that the correct classification rate can reach as high as 93%, which is quite satisfactory. When different system models are selected for the extended Kalman filter, the classification results are relatively stable. It is shown that for this particular case the result using extended Kalman filter is better than using principal component analysis.     

Progress in spectral studies of the FBS blue stellar objects

Spectral observations of 10 FBS blue stellar objects (BSO) with the OHP 1.93-m and BAO 2.6.m telescopes are reported, and overall progress in all the spectral observations and classifications of FBS BSO over 1987–2000 and in the classification of these objects based on all the accessible sources-- in all, 753 out of 1103 objects-- is discussed. Representative slit spectra for the major types of objects are presented and compared with digitized low dispersion spectra from the DFBS. The nature of the FBS objects is examined in terms of advances in the spectral studies. Two-color diagrams are constructed from the SDSS data and are used to find the regions occupied by the various types for further identification of objects of unknown type. Translated from Astrofizika, Vol. 52, No. 1, pp. 85–97 (February 2009).     

Stellar Population Analysis of Galaxies based on Genetic Algorithms

We present a new method for determining the age and relative contribution of different stellar populations in galaxies based on the genetic algorithm. We apply this method to the barred spiral galaxy NGC 3384, using CCD images in U, B, V, R and I bands. This analysis indicates that the galaxy NGC 3384 is mainly inhabited by old stellar population (age >109yr). Some problems were encountered when numerical simulations are used for determining the contribution of different stellar populations in the integrated color of a galaxy. The results show that the proposed genetic algorithm can search efficiently through the very large space of the possible ages.     

Stellar Spectral Classification Based on Capsule Network

The rapid development of large-scale sky survey project has produced a large amount of stellar spectral data, which make the automatic classification of stellar spectral data a challenging task. In this paper, we have proposed a stellar spectral classification method based on a capsule network. At first, by using the one-dimensional convolutional network and short-time Fourier transform (STFT), the one-dimensional spectra of the F5, G5, and K5 types selected from the LAMOST Data Release 5 (DR5) are converted into the two-dimensional Fourier spectrum images. Then, the two-dimensional Fourier spectrum images are classified automatically by the capsule network. Because the capsule network can preserve the hierarchical pose relationships among the entities in the image, and it does not need any pooling layers, the experimental results show that the capsule network has a better classification performance, for the classifications of the F5, G5, and K5-type stellar spectra, its classification accuracy is superior to other classification methods.     

A Study of Binary Stellar Population Synthesis of Elliptical Galaxies

We determined the relative stellar ages and metallicities of about 80 elliptical galaxies in both low and high density environments using the latest binary stellar population (BSP) synthesis model and tested the predictions of a recent hierarchical formation model that adopted the new ACDM cosmology. The stellar ages and metallicities were estimated from two high-quality published spectra line indices, the Hβand [MgFe] indices. The results show that the stellar populations of elliptical galaxies are older than 3.9 Gyr and more metal rich than 0.02. Most of our results are in agreement with the model predictions: (1) elliptical galaxies in denser environment are redder and have older populations than field galaxies; (2) elliptical galaxies with more massive stellar components are redder and have older and more metal rich populations than less massive ones; (3) the most massive galaxies have the oldest and most metal rich stars. However, some of our results differ from the model predictions on the metallicity distributions of low- and high-density elliptical galaxies and the dependence on the distance to the cluster center.