多任务Lasso回归法在恒星光谱物理参量估计中的应用

多任务学习方法在机器学习、计算机视觉、人工智能领域已得到广泛关注,利用任务间的相关性,将多个任务同时学习的效果优于每个任务单独学习的情况.采用多任务Lasso回归法(Multi-task Lasso Regression)用于恒星光谱物理参量的估计,不仅可以获取不同物理参量间的共同的特征信息,而且也可以很好地保留不同物理参量的特有的补充信息.使用恒星大气模拟模型合成光谱库ELODIE中的光谱数据和美国大型巡天项目Sloan发布的SDSS实测光谱数据进行实验,模型估算精度优于相关文献中的方法,特别是对重力加速度(lg g)和化学丰度([Fe/H])的估计.实验中通过改变光谱的分辨率,施加不同信噪比(SNR)的噪声,来说明模型的稳定性强.结果表明,模型精度受光谱分辨率和噪声的影响,但噪声对其影响更大,可见,多任务Lasso回归法不仅操作简便,稳定性强,而且也提高了模型的整体预测精度.     

CDFS天区中喑蓝星系的测光红移研究

从COMBO-17数字巡天数据里,选择了CDFS(Chandra Deep Field South)天区中1231个测光红移在0.1～0.3之间的暗蓝星系作为样本,研究了这些星系分别在只有光学波段和光学加近红外波段数据情况下做测光红移得到的红移分布,以及这些星系在静止参考系下的能谱分布(Spectral Energy Distributions,SEDs)特征.结果表明有183个星系在利用光学加近红外波段数据做测光红移时得到的红移大于1.2,它们的误差为0.046,提高测光的信噪比也有利于区分这类被光学波段误认为低红移的星系.这些暗蓝星系中高红移星系的观测近红外流量相对于光学流量有上升的趋势,而低红移星系的观测近红外流量相对于光学流量有下降的趋势.     

广义回归神经网络在日长变化预报中的应用

传统的日长变化预报多是基于线性模型,如最小二乘模型、自回归模型等,但是日长变化包含了复杂的非线性因素,线性模型预报的效果往往不甚理想.所以尝试使用一种非线性神经网络—广义回归神经网络(GRNN)模型进行日长变化预报,并将结果与使用BP (Back Propagation)神经网络模型和其它模型的预报结果进行比较.结果表明,GRNN用于日长变化预报是高效可行的.     

GALILEO搜救定位处理的分布式数据融合方法

对遇险目标搜索救援的支持是GALILEO全球卫星导航系统的一项重要民用服务.文章介绍了GALILEO搜索救援子系统(SAR)定位的基本原理.在GALILEO SAR中测量量为时间信号与频率信号,文章详细地给出了两种测量信号的定位方法.基于多传感器信息融合技术,对搜索救援系统中不同测量信号进行综合状态估计.在仿真计算中,采用带噪声的模拟测量数据进行定位,数值计算的结果显示基于多传感器数据融合方法的综合状态估计是可靠的.     

脉动变星数据库的建立

为了方便国内学者对脉动变星进行研究,建立了一个数据库,目前包含了时间长达10 yr左右的巡天项目MAssive Compact Halo Objects(大质量致密银晕天体,MACHO)和Optical Gravitational Lensing Experiment(光学引力透镜实验,OGLE)发现的脉动变星,一共容纳了来自银河系核球与大小麦哲伦云中的共近23万颗变星.采用的软件是LAMP,即Linux+Apache+MySQL+PHP.数据库的使用通过网页的简单搜索界面实现,搜索参数主要是天体的赤经、赤纬和半径.鉴于本数据库的灵活性,将来很方便加入其他的变星数据.     

类星体形成的半解析模型

以Kang等的星系形成的半解析模型为基础,研究星系和中心大质量黑洞的共同形成和演化.假设类星体的活动性由星系并合引起,通过引入黑洞吸积率,确定了红移区间0z4.5的类星体的热光度函数,并分别从爱丁顿比率、黑洞质量函数和两点相关函数3个方面对模型进行限制,模型预言的光度函数能在整个红移、整个光度范围内与观测相符合.结果显示,常数爱丁顿比率不能很好地描述黑洞吸积,爱丁顿比率需要在一定红移范围内随红移增加;星系主并合是触发类星体活动性的有效机制,而次并合只对低光度和中等光度的类星体有影响,对高光度类星体影响很小;在z=1.0处,极高光度的类星休比其他类星体具有更强的成团性.     

3C 345和3C 273的光变分析

用结构函数法,对两个经典的blazar 3C 273和3C 345的22 GHz、37 GHz和光学波段的数据进行了周期分析.结果显示,3C 273的周期性比较明显,22 GHz经分析存在7.0年和14.7年的周期,37 GHz存在7.2年和14.5年的周期,同时它的光学波段可能存在0.7年和8.7年的周期.3C 345的37 GHz存在比较明显的8.8年的周期,22 GHz和光学波段分别存在不太明显的9.3年和10.2年的周期.     

非均匀喷流模型及其在BL Lac天体中的应用

在非均匀锥形喷流模型中,电子数密度、磁场强度随着到喷流顶点的距离呈幂律分布.该模型能成功解释活动星系核喷流核心区域的平谱射电辐射,但已有的模型计算只适用于喷流运动方向与视线夹角很大的情况,所以需要建立适用于任何视角情况的非均匀锥形喷流辐射计算公式.普遍认为BL Lac天体中喷流的运动方向与视线夹角很小,推广后的非均匀喷流模型拟合了3个BL Lac天体的射电观测谱,确定了它们喷流中电子数密度、磁场强度等物理参数.研究结果表明观测辐射谱拐折频率确定出锥形喷流离黑洞最近距离,对于这3个BL Lac天体,它们的锥形喷流离黑洞最近距离约为Schwarzschild半径.     

脉冲星数据比对分析和可视化系统设计与实现

脉冲星数据比对分析和可视化系统(PSRDB,URL:http://www.psrdb.net/),由FAST(Five-hundred-meter Aperture Spherical Radio Telescope)早期科学数据中心团队为快速开展脉冲星候选体比对分析和数据管理研发.通过前端数据提交页面,接收和维护来自FAST及其他研究机构的候选体数据.目前,PSRDB已收录自1967年人类发现第1颗脉冲星以来所有公开文献发表的2811颗脉冲星样本,并采集了当前主要巡天项目尚未正式发表的源和候选体,如FAST多科学目标同时扫描巡天(CRAFTS)候选体数据.基于入库基础数据,利用位置、周期、色散等参数进行比对分析,辅助科研工作者在线检索匹配已知星表数据,最后将检索匹配、比对分析结果生成图表供进一步分析.目前,PSRDB已被应用于FAST脉冲星搜寻和候选体数据管理.未来,PSRDB可在新源认证、后随观测、观测计划制定和原始数据处理流程设计等方面提供数据和工具支撑.     

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.     

VOFilter： Bridging Virtual Observatory and Industrial Office Applications

VOFilter is an XML based filter developed by the Chinese Virtual Observatory project to transform tabular data files from VOTable format into OpenDocument format. VOTable is an XML format defined for the exchange of tabular data in the context of the Virtual Observatory (VO). It is the first Proposed Recommendation defined by International Virtual Observatory Alliance, and has obtained wide support from both the VO community and many Astronomy projects. OpenOffice.org is a mature, open source, and front office application suite with the advantage of native support of industrial standard OpenDocument XML file format. Using the VOFilter, VOTable files can be loaded in OpenOffice.org Calc, a spreadsheet application, and then displayed and analyzed as other spreadsheet files. Here, the VOFilter acts as a connector, bridging the coming VO with current industrial office applications. We introduce Virtual Observatory and technical background of the VOFilter. Its workflow, installation and usage are presented. Existing problems and limitations are also discussed together with the future development plans.     

Correlations among the properties of galaxies found in a blind H i survey, which also have SDSS optical data

We have used the Parkes Multibeam system and the Sloan Digital Sky Survey to assemble a sample of 195 galaxies selected originally from their H  i signature to avoid biases against unevolved or low surface brightness objects. For each source nine intrinsic properties are measured homogeneously, as well as inclination and an optical spectrum. The sample, which should be almost entirely free of either misidentification or confusion, includes a wide diversity of galaxies ranging from inchoate, low surface brightness dwarfs to giant spirals. Despite this diversity there are five clear correlations among their properties. They include a common dynamical mass-to-light ratio within their optical radii, a correlation between surface brightness and luminosity and a common H  i surface density. Such correlation should provide strong constrains on models of galaxy formation and evolution.     

MoMaF: the Mock Map Facility

We present the Mock Map Facility, a powerful tool for converting theoretical outputs of hierarchical galaxy formation models into catalogues of virtual observations. The general principle is straightforward: mock observing cones can be generated using semi-analytically post-processed snapshots of cosmological N -body simulations. These cones can then be projected to synthesize mock sky images. To this end, the paper describes in detail an efficient technique for creating such mock cones and images from the galaxies in cosmological simulations ( galics ) semi-analytic model, providing the reader with an accurate quantification of the artefacts it introduces at every step. We show that replication effects introduce a negative bias on the clustering signal – typically peaking at less than 10 per cent around the correlation length. We also thoroughly discuss how the clustering signal is affected by finite-volume effects, and show that it vanishes at scales larger than approximately one-tenth of the simulation box size. For the purpose of analysing our method, we show that number counts and redshift distributions obtained with galics / momaf compare well with K -band observations and the two-degree field galaxy redshift survey. Given finite-volume effects, we also show that the model can reproduce the automatic plate measuring machine angular correlation function. The momaf results discussed here are made publicly available to the astronomical community through a public data base. Moreover, a user-friendly Web interface ( http://galics.iap.fr ) allows any user to recover her/his own favourite galaxy samples through simple SQL queries. The flexibility of this tool should permit a variety of uses ranging from extensive comparisons between real observations and those predicted by hierarchical models of galaxy formation, to the preparation of observing strategies for deep surveys and tests of data processing pipelines.     

The Holmberg effect of Luminous Red Galaxy (LRG) pairs of the SDSS Data Release 4 (SDSS4)

We investigate the Holmberg effect of Luminous Red Galaxy (LRG) pairs of the SDSS Data Release 4 (SDSS4). It is found that the Holmberg effect of LRG pairs has no statistical significance. According to types of the two components in each pair, we divide the isolated LRG pair sample into three subsamples: the early+early subsample, the late+late subsample and the early+late subsample. Further analyses show that all LRG pair subsamples do not exhibit statistically significant Holmberg effect. (© 2007 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)