流形判别分析和支持向量机的恒星光谱数据自动分类方法
发布时间：2019-03-14 21:15
【摘要】：尽管经典的分类方法支持向量机SVM在天文学领域广泛应用,但其只考虑类间的绝对间隔而忽略类内的分布性状,因而分类性能有待于进一步提升。鉴于此,提出一种新颖的基于流形判别分析和支持向量机的恒星光谱数据自动分类方法。该方法引入流形判别分析的两个重要概念:基于流形的类内离散度MW和基于流形的类间离散度MB。所提方法找到的分类面同时保证MW最小且MB最大。可建立相应最优化问题,然后将原最优化问题转化为QP对偶形式求得支持向量和判别函数,最后利用判别函数判断测试样本的类属。该方法的最大优势在于进行分类决策时,不仅考虑样本的类间信息和分布特征,而且还保持了各类的局部流形结构。SDSS恒星光谱数据上的比较实验表明该方法的有效性。
[Abstract]:Although the classical classification method, support Vector Machine (SVM), is widely used in astronomy, its classification performance needs to be further improved because it only considers the absolute interval between classes but neglects the intra-class distribution characteristics. In view of this, a novel classification method of star spectral data based on manifold discriminant analysis and support vector machine is proposed. In this method, two important concepts of manifold discriminant analysis are introduced: within-class dispersion MW based on manifolds and MB. between classes based on manifolds. The classification surface found by the proposed method ensures that the MW is minimum and the MB is the largest at the same time. The corresponding optimization problem can be established, then the original optimization problem is transformed into the dual form of QP to obtain the support vector and discriminant function. Finally, the classification of the test sample is judged by using the discriminant function. The greatest advantage of the proposed method is that it not only considers the inter-class information and distribution characteristics of the samples, but also maintains the local manifold structure. The comparative experiments on the star spectral data of SDSS show the effectiveness of the proposed method.
【作者单位】： 中北大学计算机与控制工程学院;中北大学信息与通信工程学院;山西大学商务学院信息学院;
【基金】：国家自然科学基金项目(61202311) 山西省高等学校科技创新项目(20131112)资助
【分类号】：O433.4;P144.1
本文编号：2440375
[Abstract]:Although the classical classification method, support Vector Machine (SVM), is widely used in astronomy, its classification performance needs to be further improved because it only considers the absolute interval between classes but neglects the intra-class distribution characteristics. In view of this, a novel classification method of star spectral data based on manifold discriminant analysis and support vector machine is proposed. In this method, two important concepts of manifold discriminant analysis are introduced: within-class dispersion MW based on manifolds and MB. between classes based on manifolds. The classification surface found by the proposed method ensures that the MW is minimum and the MB is the largest at the same time. The corresponding optimization problem can be established, then the original optimization problem is transformed into the dual form of QP to obtain the support vector and discriminant function. Finally, the classification of the test sample is judged by using the discriminant function. The greatest advantage of the proposed method is that it not only considers the inter-class information and distribution characteristics of the samples, but also maintains the local manifold structure. The comparative experiments on the star spectral data of SDSS show the effectiveness of the proposed method.
【作者单位】： 中北大学计算机与控制工程学院;中北大学信息与通信工程学院;山西大学商务学院信息学院;
【基金】：国家自然科学基金项目(61202311) 山西省高等学校科技创新项目(20131112)资助
【分类号】：O433.4;P144.1
【相似文献】
中国期刊全文数据库 前10条
1 赵水兵;严冰;樊书宏;刘晓东;;水下目标空间源数特征提取与识别方法[J];声学技术;2008年03期
2 ;[J];;年期
3 ;[J];;年期
4 ;[J];;年期
5 ;[J];;年期
6 ;[J];;年期
7 ;[J];;年期
8 ;[J];;年期
9 ;[J];;年期
10 ;[J];;年期
，本文编号：2440375
本文链接：https://www.wllwen.com/kejilunwen/tianwen/2440375.html
