CASIA OpenIR  > 毕业生  > 博士学位论文
Alternative TitleStudies on automated spectral recognition of celestial objects
Thesis Advisor胡占义 ; 赵永恒
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword天体光谱 自动识别 Pca Svms Smm-rbfnn Celestial Object Spectrum Automated Recognition Principal Component Analysis Support Vector Machines Statistical Mixture Modelin
Abstract在我国建造的LAMOST望远镜建成后,每晚将有2-4万条光谱需要进行 自动的分类识别及参数测量,因此急需研究相应的技术和算法。本文针对天体 光谱的自动识别问题进行了研究。结合已有的技术,与天文学家共同设计了自 动识别与分析的流程图。并针对关键的粗分类环节的不足,进行低信噪比光谱 的自动识别研究。对恒星细分类问题,针对快速性和流量未定标情况的识别需 求,进行自动识别研究。论文的主要工作有以下五个方面: (1)NOs与ELOs的自动识别研究 根据光谱是吸收型或发射型,可以把天体分成正常天体(NOs)与发射线 天体(ELOs)。由于吸收、发射型光谱的通常判断方法(通过判别吸收或发射 特征谱线的多寡确定吸收型或发射型光谱)对信噪比较低时的光谱效果不好, 因此,我们提出基于PCA三维特征空间及SVMs算法寻找分类面的识别方法。 该方法通过对河外天体光谱进行红移模拟,得到红移光谱较全的样本集,并采 用PCA方法提取三维特征矢量,为SVMs分类面寻找提高了速度。另外,SVMs 分类面能有好的推广性能。通过实验表明,此方法具有训练和识别速度快,精 度高,鲁棒性强的特点。 (2)恒星与NGs的自动识别研究 在红移值未知的时候,恒星与正常星系(NGs)的自动识别,可以通过判 断6563 A处附近是否有吸收线来确定是恒星,但在低信噪比光谱中,由于噪声 的影响,容易造成“真”谱线提取不出来,而提取出来的有“假”谱线。为此, 我们提出了基于SMM-RBFNN的恒星与NGs识别方法。该方法先采用PCA对 红移模拟的NGs样本和恒星一起进行特征提取。在50维的特征空间中,采用 组合径向基网络——SMM-RBFNN作为识别器。由于SMM-RBFNN是对RBF 网络的识别性能采用混合概率模型建模,并采用类EM算法对各模块的参数进 行协同训练,因此,它的识别性能比单个的RBF网络效果好。实验证明,该方 法的训练速度快,识别精度高,鲁棒性强,适合于低信噪比下的光谱识别。 (3)NGs与AGs的自动识别研究 NGs与AGs的识别研究是第(1)问题的子问题,我们通过PCA+ODP方 法,可以找到各种红移下的两类光谱按类在ODP平面呈现规律的分布,因此作 为单独的内容进行讨论。采用ODP识别器在实验中获得了较高的识别精度。该 方法为深入研究天体光谱不随红移改变的类型分布奠定了基础。 (4)恒星的光谱型自动识别研究针对已有神经网络技术中的速度问题和巡天观测中的光谱流量未定标情 况,提出了如下方法:a.基于PCA的二
Other AbstractAfter the scheduled completion of LAMOST project in China at the end of 2004, about 20,000 to 40,000 spectra will be collected at each observation night,such voluminous data urgently demand to explore automatic recognition and parameters measurement methods consequently.This thesis is focused on automated recognition methods of celestial objects via their spectra.Since it is expected that the collected spectra in LAMOST will be of low SNR,and of uncalibrated flux in part of targes, the special attention is also dedicated to deal with such problems in our work.The main points of our work can be summarized as follows: (1)The study on automated recognition between normal objects(NOs)and emission-line objects(ELOs). The celestial objects are classified as normal and emission-line objects depending on whether their spectra are of absorption or emission type.The traditional method for the classification is to estimate the respective number of absorption and emission lines emerged in the spectra,its performance usually degrades rapidly with low SNR data.In this work,a PCA+SVMs method is proposed for the classification.Firstly, the first three principal components in the PCA are used to extract features,then SVM is followed to classify an incoming spectrum as either an NO or an ELO.In order to offset the shortage of spectra under various redshifts,simulated data perturbed by Gaussian noise under different noise level are generated during both the training and verification stages.Experiments show that the proposed method is both efficient and robust. (2)The study o f automated recognition between stars and normal galaxies(NGs) In case the spectra redshifts are unknown,the stars and NGs can usually be distinguished by locating the absorption line on 6563 A.However.when the SNR of spectra is low,the absorption line could be submerged by noise.In this work,an alternative method is proposed,which is based on statistical mixture modeling with RBF neural networks,in short SMM-RBFNN.In this method,a 50-dimensional feature space is created by PCA firstly, then the SMM-RBFNN is used for the final classification. Since SMM-RBFNN is a statistical mixture model of RBFNNS and the parameters of different modules are trained concurrently with an EM-like algorithm,it has some inherent advantages over an individual RBFNN.Experiments show that the proposed method is of high computational efficiency, good accuracy and strong robustness. (3)The study of automated recognition between normal and active galaxies(AGs) A PCA+ODP(Optimal discrimination plane)method is used to separate the NGs and AGs.It is shown that the extracted features by PCA from NGs and AGs can well be clustered separately on the ODP obtained by the Fisher Discrimination Criterion. Besides,this discrimination is undertaken with unknown redshifts. (4)The study of automated recognition of spectral class from stellar spectra Two methods are proposed for stellar re
Other Identifier728
Document Type学位论文
Recommended Citation
GB/T 7714
覃冬梅. 天体光谱信号的自动识别方法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2003.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[覃冬梅]'s Articles
Baidu academic
Similar articles in Baidu academic
[覃冬梅]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[覃冬梅]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.