基于Stacking集成学习的恒星/星系分类研究

机器学习在当今诸多领域已经取得了巨大的成功,但是机器学习的预测效果往往依赖于具体问题.集成学习通过综合多个基分类器来预测结果,因此,其适应各种场景的能力较强,分类准确率较高.基于斯隆数字巡天(Sloan Digital Sky Survey,SDSS)计划恒星/星系中最暗源星等集分类正确率低的问题,提出一种基于Stacking集成学习的恒星/星系分类算法.从SDSS-DR7(SDSS Data Release 7)中获取完整的测光数据集,并根据星等值划分为亮源星等集、暗源星等集和最暗源星等集.仅针对分类较为复杂且困难的最暗源星等集展开分类研究.首先,对最暗源星等集使用10折嵌套交叉验证,然后使用支持向量机(Support Vector Machine,SVM)、随机森林(Random Forest,RF)、XGBoost(eXtreme Gradient Boosting)等算法建立基分类器模型;使用梯度提升树(Gradient Boosting Decision Tree,GBDT)作为元分类器模型.最后,使用基于星系的分类正确率等指标,与功能树(Function Tree,FT)、SVM、RF、GBDT、XGBoost、堆叠降噪自编码(Stacked Denoising AutoEncoders,SDAE)、深度置信网络(Deep Belief Network,DBN)、深度感知决策树(Deep Perception Decision Tree,DPDT)等模型进行分类结果对比分析.实验结果表明,Stacking集成学习模型在最暗源星等集分类中要比FT算法的星系分类正确率提高了将近10%.同其他传统的机器学习算法、较强的提升算法、深度学习算法相比,Stacking集成学习模型也有较大的提升.     

Study of Star/Galaxy Classification Based on the XGBoost Algorithm

Machine learning has achieved great success in many areas today. The lifting algorithm has a strong ability to adapt to various scenarios with a high accuracy, and has played a great role in many fields. But in astronomy, the application of lifting algorithms is still rare. In response to the low classification accuracy of the dark star/galaxy source set in the Sloan Digital Sky Survey (SDSS), a new research result of machine learning, eXtreme Gradient Boosting (XGBoost), has been introduced. The complete photometric data set is obtained from the SDSS-DR7, and divided into a bright source set and a dark source set according to the star magnitude. Firstly, the ten-fold cross-validation method is used for the bright source set and the dark source set respectively, and the XGBoost algorithm is used to establish the star/galaxy classification model. Then, the grid search and other methods are used to adjust the XGBoost parameters. Finally, based on the galaxy classification accuracy and other indicators, the classification results are analyzed, by comparing with the models of function tree (FT), Adaptive boosting (Adaboost), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Stacked Denoising AutoEncoders (SDAE), and Deep Belief Nets (DBN). The experimental results show that, the XGBoost improves the classification accuracy of galaxies in the dark source classification by nearly 10% as compared to the function tree algorithm, and improves the classification accuracy of sources with the darkest magnitudes in the dark source set by nearly 5% as compared to the function tree algorithm. Compared with other traditional machine learning algorithms and deep neural networks, the XGBoost also has different degrees of improvement.     

Correlations Between Color Gradients and Structural Parameters of Disks in Late-type Spiral Galaxies

From the second group of the data released by the Sloan numerical patrol survey (SDSS DR2), 395 face-on late-type spiral galaxies brighter than 15-th magnitude in the waveband r are selected as the sample. The relations between the color gradients and structural parameters of disks are studied. The results indicate the following: (1) The color gradient is independent of the absolute magnitude (mass) of the disk. (2) The color gradient of disk is concerned with the size of disk. The larger the disk, the steeper the color gradient. (3) The color gradient of disk is concerned with the color of disk. The bluer the disk color, the steeper the gradient. (4) The color gradient of disk is related with the surface brightness of disk. The brighter the disk, the steeper the color gradient. Moreover, the historical constraint formed by the relations between the color gradients of disks and various structural parameters is briefly discussed.     

Structure through colour: a pixel approach towards understanding galaxies

We present a study of pixel colour–magnitude diagrams (pCMDs) for a sample of 69 nearby galaxies chosen to span a wide range of Hubble types. Our goal is to determine how useful a pixel approach is for studying galaxies according to their stellar light distributions and content. The galaxy images were analysed on a pixel-by-pixel basis to reveal the structure of the individual pCMDs. We find that the average surface brightness (or projected mass density) in each pixel varies according to galaxy type. Early-type galaxies exhibit a clear 'prime sequence' and some pCMDs of face-on spirals reveal 'inverse-L' structures. We find that the colour dispersion at a given magnitude is found to be approximately constant in early-type galaxies but this quantity varies in the mid and late types. We investigate individual galaxies and find that the pCMDs can be used to pick out morphological features. We discuss the discovery of 'Red Hooks' in the pCMDs of six early-type galaxies and two spirals and postulate their origins. We develop quantitative methods to characterize the pCMDs, including measures of the blue-to-red light ratio and colour distributions of each galaxy and we organize these by morphological type. We compare the colours of the pixels in each galaxy with the stellar population models of Bruzual & Charlot to calculate star formation histories for each galaxy type and compare these to the stellar mass within each pixel. Maps of pixel stellar mass and mass-to-light ratio are compared to galaxy images. We apply the pCMD technique to three galaxies in the Hubble Ultra Deep Field to test the usefulness of the analysis at high redshift. We propose that these results can be used as part of a new system of automated classification of galaxies that can be applied at high redshift.     

Environmental dependence of five photometric band concentration indexes for luminous red galaxies from the SDSS Data Release 7

Using the Luminous Red Galaxy (LRG) sample of the Sloan Digital Sky Survey Data Release 7 (SDSS DR7), we explore the environmental dependence of five photometric band concentration indexes of LRGs. It turns out that the trends for all the five band concentration indexes to change with the local density are the same: galaxies with higher concentration reside preferentially in overdense environments, while galaxies with lower concentration are located preferentially in underdense environments (© 2010 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)     

Evolution of the u-band luminosity function from redshift 1.2 to 0

We produce and analyse u -band (  λ≈ 355  nm) luminosity functions (LFs) for the red and blue populations of galaxies using data from the Sloan Digital Sky Survey (SDSS) u -band Galaxy Survey ( u GS) and Deep Evolutionary Exploratory Probe 2 (DEEP2) survey. From a spectroscopic sample of 41 575 SDSS u GS galaxies and 24 561 DEEP2 galaxies, we produce colour magnitude diagrams and make use of the colour bimodality of galaxies to separate red and blue populations. LFs for eight redshift slices in the range  0.01 < z < 1.2  are determined using the  1/ V _max  method and fitted with Schechter functions showing that there is significant evolution in   M *  , with a brightening of 1.4 mag for the combined population. The integration of the Schechter functions yields the evolution in the u -band luminosity density (LD) out to   z ∼ 1  . By parametrizing the evolution as  ρ∝ (1 + z )^β  , we find that  β= 1.36 ± 0.2  for the combined populations and  β= 2.09 ± 0.2  for the blue population. By removing the contribution of the old stellar population to the u -band LD and correcting for dust attenuation, we estimate the evolution in the star formation rate (SFR) of the Universe to be  β_SFR= 2.5 ± 0.3  . Discrepancies between our result and higher evolution rates measured using the infrared and far-UV can be reconciled by considering possibilities such as an underestimated dust correction at high redshifts or evolution in the stellar initial mass function.     

Point spread function tails and the measurements of diffuse stellar halo light around edge-on disc galaxies

Measuring the integrated stellar halo light around galaxies is very challenging. The surface brightness of these haloes is expected to be many magnitudes below dark sky and the central brightness of the galaxy. Here, I show that in some of the recent literature the effect of very extended Point Spread Function (PSF) tails on the measurements of halo light has been underestimated; especially in the case of edge-on disc galaxies. The detection of a halo along the minor axis of an edge-on galaxy in the Hubble Ultra Deep Field can largely be explained by scattered galaxy light. Similarly, depending on filter and the shape one assumes for the uncertain extended PSF, 20–80 per cent of the halo light found along the minor axis of scaled and stacked Sloan Digital Sky Survey (SDSS) edge-on galaxy images can be explained by scattered galaxy light. Scattered light also significantly contributes to the anomalous halo colours of stacked SDSS images. The scattered light fraction decreases when looking in the quadrants away from the minor axis. The remaining excess light is well modelled with a Sérsic profile halo with shape parameters based on star count halo detections of nearby galaxies. Even though, the contribution from PSF scattered light does not fully remove the need for extended components around these edge-on galaxies, it will be very challenging to make accurate halo light shape and colour measurements from integrated light without very careful PSF measurements and scattered light modelling.     

A study of active galactic nuclei in low surface brightness galaxies with Sloan Digital Sky Survey spectroscopy

Active galactic nuclei (AGNs) in low surface brightness galaxies (LSBGs) have received little attention in previous studies. We present a detailed spectral analysis of 194 LSBGs from the Impey et al. (1996) APM LSBG sample which has been observed spec-troscopically by the Sloan Digital Sky Survey Data Release 5 (SDSS DR5). Our elaborate spectral analysis enables us to carry out, for the first time, reliable spectral classification of nuclear processes in LSBGs based on the standard emission line diagnostic diagrams in a rigorous way. Star-forming galaxies are common, as found in about 52% of LSBGs. We find that, contrary to some previous claims, the fraction of galaxies that contain AGNs is significantly lower than that found in nearby normal galaxies of high surface brightness. This is qualitatively in line with the finding of Impey et al. This result holds true even within each morphological type from Sa to Sc. LSBGs that have larger central stellar ve-locity dispersions or larger physical sizes tend to have a higher chance of harboring an AGN. For three AGNs with broad emission lines, the black hole masses estimated from the emission lines are broadly consistent with the well known M-σ* relation established for normal galaxies and AGNs.     

晚型旋涡星系盘的颜色梯度与结构参数的相关关系

从Sloan数字巡天第2批释放的数据(SDSS DR2)中选择了395个在r波段亮于15等的面向晚型旋涡星系作为样本,研究了盘的颜色梯度与结构参数的关系.结果表明:盘的颜色梯度与盘的绝对星等(质量)无关;盘的颜色梯度与盘的尺度有关,越大的盘颜色梯度越陡;盘的颜色梯度与盘的颜色有关,越蓝的盘颜色梯度越陡;盘的颜色梯度与盘的表面亮度有关,越亮的盘颜色梯度越陡,并简单讨论了盘的颜色梯度与各结构参数的相关关系对晚型旋涡星系盘恒星形成历史的约束.     

The star formation histories of galaxies in the Sloan Digital Sky Survey

We present the results of a moped analysis of  ∼3 × 10⁵  galaxy spectra from the Sloan Digital Sky Survey Data Release 3 (SDSS DR3), with a number of improvements in data, modelling and analysis compared with our previous analysis of DR1. The improvements include: modelling the galaxies with theoretical models at a higher spectral resolution of 3 Å, better calibrated data, an extended list of excluded emission lines and a wider range of dust models. We present new estimates of the cosmic star formation rate (SFR), the evolution of stellar mass density and the stellar mass function from the fossil record. In contrast to our earlier work the results show no conclusive peak in the SFR out to a redshift around 2 but continue to show conclusive evidence for 'downsizing' in the SDSS fossil record. The star formation history is now in good agreement with more traditional instantaneous measures. The galaxy stellar mass function is determined over five decades of mass, and an updated estimate of the current stellar mass density is presented. We also investigate the systematic effects of changes in the stellar population modelling, the spectral resolution, dust modelling, sky lines, spectral resolution and the change of data set. We find that the main changes in the results are due to the improvements in the calibration of the SDSS data, changes in the initial mass function and the theoretical models used.     

基于NPSVM模型的超新星识别方法

巡天观测与高能物理、黑洞天文等领域均有密切的联系.基于星系-超新星二分类问题,研究光谱数据预处理,结合余弦相似度改善PCA(Principal Component Analysis)光谱分解特征提取方法,用SDSS(the Sloan Digital Sky Survey)、WISeREP(the Weizmann Interactive Supernova data REPository)组成的5620条光谱数据集训练支持向量机,可以得到0.498%泛化误差的识别模型和新样本分类概率.使用Neyman-Pearson决策方法建立NPSVM(Neyman-Pearson Support Vector Machine)模型可进一步降低超新星的漏判率.     

Automated Separation of Stars and Normal Galaxies Based on Statistical Mixture Modeling with RBF Neural Networks

For LAMOST, the largest sky survey program in China, the solution of the problem of automatic discrimination of stars from galaxies by spectra has shown that the results of the PSF test can be significantly refined. However, the problem is made worse when the redshifts of galaxies are not available. We present a new automatic method of star/(normal) galaxy separation, which is based on Statistical Mixture Modeling with Radial Basis Function Neural Networks (SMM-RBFNN). This work is a continuation of our previous one, where active and non-active celestial objects were successfully segregated. By combining the method in this paper and the previous one, stars can now be effectively separated from galaxies and AGNs by their spectra-a major goal of LAMOST, and an indispensable step in any automatic spectrum classification system. In our work, the training set includes standard stellar spectra from Jacoby's spectrum library and simulated galaxy spectra of EO, SO, Sa, Sb types with redshift ranging from 0 to 1     

Stellar spectra classification with twin hypersphere model

With the increase of stellar spectra, how to automatically classify these spectra have attracted astronomer's attention. Support Vector Machine (SVM), as a typical classifier, has widely used in stellar spectra classification. Due to its limited performance in various classification problems and higher training time, a model with a pair of hyperspheres named Twin Hypersphere Model (THM), proposed by Peng and Xu, is utilized for stellar spectra classification in this paper. In THM, the samples in one hypersphere is far from another according to the Euclidean distance. The comparative experiments with SVM and Twin Support Vector Machine (TWSVM) on the SDSS datasets shows that the THM model gives the best classification accuracy of 0.8836 for type F, 0.9446 for type G, and 0.9509 for type K, which are better than the classification accuracies of 0.8000, 0.8484, 0.8911 obtained by SVM and 0.8413, 0.8699, 0.9109 obtained by TWSVM. It can be concluded that THM perform better than traditional techniques such as SVM and TWSVM on the K-, F-, G- type stellar spectra classification.     

Restrictions to the galaxy evolutionary models from the Hawaiian Deep Fields SSA13 and SSA22

Quantitative structural analysis of the galaxies present in the Hawaiian Deep Fields SSA13 and SSA22 is reported. The structural parameters of the galaxies have been obtained automatically by fitting a two-component model (Sérsic r ^{1/ n} bulge and exponential disc) to the surface brightness of the galaxies. The galaxies were classified on the basis of the bulge-to-total luminosity ratio  ( B / T )  . The magnitude selection criteria and the reliability of our method have been checked by using Monte Carlo simulations. A complete sample of objects up to redshift 0.8 has been achieved. Spheroidal objects (E/S0) represent ≈33 per cent and spirals ≈41 per cent of the total number of galaxies, while mergers and unclassified objects represent ≈26 per cent. We have computed the comoving space density of the different kinds of object. In an Einstein–de Sitter universe, a decrease in the comoving density of E/S0 galaxies is observed as redshift increases (≈30 per cent less at   z =0.8)  , while for spiral galaxies a relatively quiet evolution is reported. The framework of hierarchical clustering evolution models of galaxies seems to be the most appropriate to explain our results.     

Environmental dependence of stellar mass,SFR, and SSFR for the Luminous Red Galaxy (LRG) sample of the SDSS DR7

Using the Luminous Red Galaxy (LRG) sample of the Sloan Digital Sky Survey Data Release 7 (SDSS DR7), we investigate the environmental dependence of stellar mass, star formation rate (SFR) and specific star formation rate (SSFR) of LRGs. It is found that stellar mass of LRGs nearly is independent of local environments, and that the environmental dependence of SFR and SSFR in the LRG sample is much weaker than the one in the Main galaxy sample. One possible explanation is that galaxy color and morphology are a pair of galaxy properties most predictive of local environments, while LRGs are a group of galaxies that are likely to be luminous, red and of early types (© 2012 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)     

The shapes of galaxies in the Sloan Digital Sky Survey

We determine the underlying shapes of spiral and elliptical galaxies in the Sloan Digital Sky Survey Data Release 6 (SDSS DR6) from the observed distribution of projected galaxy shapes, taking into account the effects of dust extinction and reddening. We assume that the underlying shapes of spirals and ellipticals are well approximated by triaxial ellipsoids. The elliptical galaxy data are consistent with oblate spheroids, with a correlation between luminosity and ellipticity: the mean values of minor to middle axis ratios are 0.41 ± 0.03 for   M _r ≈−18  ellipticals and 0.76 ± 0.04 for   M _r ≈−22.5  ellipticals. Ellipticals show almost no dependence of axial ratio on galaxy colour, implying a negligible dust optical depth.
There is a strong variation of spiral galaxy shapes with colour indicating the presence of dust. The intrinsic shapes of spiral galaxies in the SDSS DR6 are consistent with flat discs with a mean and dispersion of thickness to diameter ratio of (21 ± 2) per cent, and a face-on ellipticity, e , of  ln( e ) =−2.33 ± 0.79  . Not including the effects of dust in the model leads to discs that are systematically rounder by up to 60 per cent. More luminous spiral galaxies tend to have thicker and rounder discs than lower luminosity spirals. Both elliptical and spiral galaxies tend to be rounder for larger galaxies.
The marginalized value of the edge-on r -band dust extinction E ₀ in spiral galaxies is   E ₀≃ 0.45  mag for galaxies of median colours, increasing to   E ₀= 1  mag for   g − r > 0.9  and   E ₀= 1.9  for the luminous and most compact galaxies, with half-light radii  <2  h ⁻¹ kpc  .     

Recovering physical parameters from galaxy spectra using MOPED

We derive physical parameters of galaxies from their observed spectra using MOPED, the optimized data compression algorithm of Heavens, Jimenez & Lahav. Here we concentrate on parametrizing galaxy properties, and apply the method to the NGC galaxies in Kennicutt's spectral atlas. We focus on deriving the star formation history, metallicity and dust content of galaxies. The method is very fast, taking a few seconds of CPU time to estimate ∼17 parameters, and is therefore specially suited to studying large data sets, such as the Anglo-Australian two-degree-field (2dF) galaxy survey and the Sloan Digital Sky Survey (SDSS). Without the power of MOPED, the recovery of star formation histories in these surveys would be impractical. In Kennicutt's atlas, we find that for the spheroidals a small recent burst of star formation is required to provide the best fit to the spectrum. There is clearly a need for theoretical stellar atmospheric models with spectral resolution better than 1 Å if we are to extract all the rich information that large redshift surveys contain in their galaxy spectra.     

Isolated Main Galaxy Pairs from the SDSS Data Release 4

From the Main galaxy data of the SDSS Data Release 4 (SDSS4), we have identified close galaxy pairs at neighbourhood radius R = 100 kpc by three-dimensional cluster analysis. Using the criterion that an "isolated galaxy pair" must be separated from its "nearest neighbor" by more than 500 kpc, we constructed an isolated galaxy pair sample of 1158 pairs. We also constructed a random pair sample by randomly selecting 1158 galaxy pairs from the Main galaxy sample, which has the same redshift distribution as the isolated galaxy pair sample, and in which the two components of any pair have the same redshifts. Comparative studies of luminosity and size between the members of the galaxy pairs are performed. We find and further confirm there is no tendency for paired galaxies to have similar luminosities or sizes. From the isolated pair sample we also selected a subsample with the magnitude limit of the primary raised by 2 magnitudes, so as to include pairs in which the secondary is 2 magnitudes fainter than the primary. This subsample contains 82 pairs. A random pair sample is similarly constructed.     

The Tully–Fisher relation for S0 galaxies

We present a study of the local B - and K _s-band Tully–Fisher relation (TFR) between absolute magnitude and maximum circular speed in S0 galaxies. To make this study, we have combined kinematic data, including a new high-quality spectral data set from the Fornax cluster, with homogeneous photometry from the Third Reference Catalogue of Bright Galaxies and Two Micron All Sky Survey catalogues, to construct the largest sample of S0 galaxies ever used in a study of the TFR. Independent of environment, S0 galaxies are found to lie systematically below the TFR for nearby spirals in both optical and infrared bands. This offset can be crudely interpreted as arising from the luminosity evolution of spiral galaxies that have faded since ceasing star formation.
However, we also find a large scatter in the TFR. We show that most of this scatter is intrinsic, not due to the observational uncertainties. The presence of such a large scatter means that the population of S0 galaxies cannot have formed exclusively by the above simple fading mechanism after all transforming at a single epoch. To better understand the complexity of the transformation mechanism, we have searched for correlations between the offset from the TFR and other properties of the galaxies such as their structural properties, central velocity dispersions and ages (as estimated from line indices). For the Fornax cluster data, the offset from the TFR correlates with the estimated age of the stars in the individual galaxies, in the sense and of the magnitude expected if S0 galaxies had passively faded since being converted from spirals. This correlation implies that a significant part of the scatter in the TFR arises from the different times at which galaxies began their transformation.