一种新的光谱特征提取方法

发布时间：2018-09-04 06:53

【摘要】：研究了天体光谱的特征提取问题,这是光谱自动处理中的一个关键环节。通过特征提取,不仅能够约简数据、减少冗余,而且亦能抑制噪声干扰,对识别系统的精度和效率均有重要影响。提出了一种基于空间转换和分解的特征分析模型(STP),基于此,可实现对常用光谱特征提取方法的分析,例如,无监督的主成分分析(PCA),小波变换(Wavelet),有监督的支持向量机(SVM),相关向量机(RVM)和线性判别分析方法(LDA)等。在STP模型中,关注的核心要素是特征提取中对数据成分的分解、重组,以及噪声的抑制和冗余的消除。亦在STP框架的基础上,给出了一种逻辑和实现均较为简单的特征提取方法:基于曲线拟合与下采样的光谱特征提取(EFCD)。研究的一个重要发现是,在一些分类问题中文献中设计巧妙的特征提取方法并不一定是决定性的:即使采用通常的信号下采样方法提取特征,亦能获得良好的光谱识别性能,而重要的仅仅是需要将特征数量保持在一定的水平以上即可。研究中,选用的测试数据是SDSS中的Galaxy和QSO两类河外天体实测光谱,他们一般具有较大的红移,在天体光谱识别中具有较强的代表性。
[Abstract]:The feature extraction of celestial spectrum is studied, which is a key step in automatic spectral processing. The feature extraction can not only reduce the data, reduce the redundancy, but also suppress the noise interference, which has an important effect on the accuracy and efficiency of the recognition system. This paper presents a feature analysis model based on spatial transformation and decomposition (STP),). Based on this model, it can be used to analyze common spectral feature extraction methods. For example, Unsupervised principal component analysis (PCA), wavelet transform (Wavelet), supervised support vector machine (SVM), correlation vector machine (RVM) and linear discriminant analysis method (LDA) et al. In the STP model, the key elements concerned are the decomposition, reorganization, noise suppression and redundancy elimination in feature extraction. On the basis of STP framework, a simple feature extraction method based on both logic and implementation is presented: spectral feature extraction based on curve fitting and downsampling for (EFCD). One of the important findings of the study is that the design of clever feature extraction methods in some literature is not necessarily decisive: even with the usual signal subsampling method, good spectral recognition performance can be obtained. The important thing is to keep the number of features above a certain level. In the study, the measured data are measured spectra of two kinds of extragalactic bodies in SDSS, Galaxy and QSO. They generally have a large redshift and are representative in the recognition of celestial body spectrum.
【作者单位】：华南师范大学数学科学学院;聊城大学东昌学院数学系;
【基金】：国家自然科学基金项目(61075033)资助
【分类号】：P141.5

【参考文献】