CASIA OpenIR
红外光谱基线校正算法研究
韩权杰
Subtype博士
Thesis Advisor彭思龙
2019-06-03
Degree Grantor中国科学院自动化研究所
Place of Conferral智能化大厦三层第五会议室
Degree Discipline计算机应用技术
Keyword红外光谱,稀疏表示,最大相关熵准则,分位数回归,增广拉格朗日优化
Abstract

红外光谱技术作为一种简单可靠的技术,在普通化学特别是有机化学中得到了广泛的应用。它可以用来检测样品的结构,以及对化合物各种成分进行分析。然而,由于测量仪器、环境以及制备样品的影响,采集的光谱通常会受到诸如基线和噪声的影响。如何消除这些因素的干扰,建立一个稳定可靠的反映样品光谱,特别是光谱谱峰与主要成分含量的校正模型是一个需要重点研究的问题。在此背景下,针对改善光谱质量,建立稳健、高精度的模型,本文研究了基线校正、谱峰拟合、去噪以及控制基线光滑性的超参数选择等方法,主要贡献有以下三个方面:

(1)针对样本的光谱由真实的光谱、基线和噪声组成,提出了基于稀疏表示的同时光谱拟合和基线校正算法。根据基线的光滑性,并且利用类佛克脱线型构造冗余字典以及稀疏表示方法来重建样本的真实光谱。由于真实光谱是非负的,对表示系数施加了非负性约束。该算法将谱峰拟合问题和基线校正问题同步起来,能够同时处理多条光谱受基线干扰的问题且能对光谱进行一定程度的去噪。仿真实验和真实数据实验结果表明,该方法能较好地估计基线和纯谱且优于其他基线校正和预处理方法。

(2)针对最小二乘法对非高斯噪声和离群值是敏感的,而最大相关熵度量能够有效地抑制它们的影响,提出了基于最大相关熵准则的光谱拟合与基线校正方法。由于最大相关熵准则关于误差中的参数是非线性以及非凸的,利用半二次优化技巧,将其转化为最大相关熵准则的一种加法和一种乘法形式求解。实验结果表明该算法相对于基于最小二乘的光谱拟合和基线校正算法,有效地提高了后续定量模型的分析效果。

(3)针对控制基线光滑性的超参数选择往往决定所估计的基线是否有效,进而影响后续定量模型的性能。而人工选取超参数带有盲目性且需要不断地实验,耗时且没有推广性。提出了利用增广拉格朗日优化的迭代重加权分位数回归基线校正算法,通过增广拉格朗日优化方法将超参数转化为算法中的一个惩罚系数,从而使得算法能够自适应地选择超参数。该算法利用B样条基函数拟合基线,且分位数回归能够在一定程度上对噪声鲁棒,比通过格搜索进行参数选择的基线校正方法能取得更好的定量效果。

Other Abstract

Infrared spectroscopy is a simple and reliable technique which has been widely used in general chemistry, especially in organic chemistry.  It can be used to detect the structure of samples and analyze the various components of compounds. However, due to the influence of measuring instruments, environment and sample preparation, the collected spectra are usually subject to influences such as baseline and noise.  How to eliminate the interference of these factors and establish a stable and reliable calibration model which reflects the spectra of samples and the concentration of main components is an urgent problem needed to be researched. To improve spectral quality and establish a robust model, this dissertation focuses on baseline correction, spectrum fitting, denoising and selection of hyperparameter which controls the smoothness of baseline. The contents in this thesis mainly contributed to the following three aspects:

(1) Based on the collected spectrum of the sample consists of the pure spectrum, baseline and associated noise, a simultaneous spectrum fitting and baseline correction using sparse representation algorithm is proposed. With smooth prior of the baseline and a redundant dictionary is constructed by the Voight-like lineshapes, then the pure spectrum is represented by the dictionary using sparse representation. Since the pure spectrum is nonnegative, the representation coefficients are also made to be nonnegative. The algorithm can simultaneously estimates the pure spectrum and baseline and deal with the problem that multiple spectra are interfered by the baselines and denoise the spectra to a certain extent. The results of quantitative analysis show that our method successfully estimates the baseline and pure spectrum and are superior to other baseline correction and preprocessing methods.

(2) Since the least squares method is sensitive to non-gaussian noise and outliers and maximum correntropy metric can effectively suppress their influence, a maximum correntropy criterion based spectrum fitting and baseline correction algorithm is presented. Due to the maximum correntropy criterion is nonlinear and non-convex with respect to the parameters in the error, half-quadratic optimization technique is deployed and the parameters are solved by an addition and a multiplication form of maximum correntropy criterion. Compared with spectrum fitting and baseline correction algorithm based on least squares, experimental results show that the proposed method can effectively improve the performance of subsequent quantitative model.

(3) Due to the selection of hyperparameter which controls the smoothness of baseline often determines whether the estimated baseline is effective or not, then affects the performance of subsequent quantitative models. Besides, manual selection of hyperparameter is blind and requires trial and error, which is time consuming and without generality. An iterative reweighted quantile regression using augmented Lagrangian optimization for baseline correction is proposed. The hyperparameter is transformed into a penalty parameter by using augmented Lagrangian optimization method and the algorithm can select the hyperparameter adaptively. The algorithm uses B-spline basis to represent the baseline and quantile regression to suppress the influence of noise on the baseline estimation. Experimental results show that the proposed method is more effective than other baseline correction methods whose parameters are chosen by grid search.

Subject Area化学
Pages112
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23865
Collection中国科学院自动化研究所
毕业生_博士学位论文
Recommended Citation
GB/T 7714
韩权杰. 红外光谱基线校正算法研究[D]. 智能化大厦三层第五会议室. 中国科学院自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
Thesis.pdf(6813KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[韩权杰]'s Articles
Baidu academic
Similar articles in Baidu academic
[韩权杰]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[韩权杰]'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.