拉曼光谱具有非接触、非破坏性、低成本、高通量优势, 受到了多组分体系分析的关注。独立成分分析(ICA)既是多元统计方法, 也是盲分离方法, 可以无需先验知识, 只需通过测量到的混合光谱就能解出体系中各组分的估计源光谱。但是当源光谱间存在显著重叠时, ICA分离结果不可靠。本文提出了一种通过对体系光谱求导、ICA分离, 逐级剔除分量后再分离的改进ICA定性分析算法(Derivation, Separation, Cullingand Separation, DSCS-ICA), 分离得到源光谱的近似估计, 实现体系定性分析, 解决了因光谱重叠导致的现有的ICA算法分离性能差的问题。依照布洛芬胶囊配方, 配制了布洛芬、硬脂酸、聚乙烯吡咯烷酮K30、淀粉和蔗糖五种组分不同比例混合的12份布洛芬胶囊样本, 并采集其拉曼光谱数据; 采用DSCS-ICA法解出分量(ICS), 并将ICS与源光谱进行比较, 以相关系数r来判断ICS与源光谱的一致性。结果表明: 与FastICA相比, DSCS-ICA效果显著改善, ICS与源光谱对应的相关系数r达到了0.99以上, 结果具有良好的可靠性和对应性。本案例可为药物处方成分的反向研究提供参考, 并可推广应用于其他多组分体系的定性分析。

Raman spectroscopy has the advantages of non-contact, non-destructive, low cost and high throughput, and has been paid more attention to in the analysis of multi-component systems. Independent component analysis (ICA) is not only a multivariate statistical method, but also a blind separation method. It can solve the estimated source spectra of each component in the system only by measuring the mixed spectra without prior knowledge. However, when there is significant overlap between the source spectra, ICA separation results are not reliable. In this paper, an improved ICA qualitative analysis algorithm is proposed, which is based on the derivation of the spectrum of the system, ICA separation, and then separation after removing components step by step (Derivation, Separation,Culling and Separation, DSCS-ICA),the approximate estimation of the source spectrum is obtained by separation, and the qualitative analysis of the system is realized, which solves the problem of poor separation performance of the existing ICA algorithm caused by spectral overlap.According to the formula of ibuprofen capsule, 12 samples of ibuprofen capsule were prepared with different proportions of five components, including ibuprofen, stearic acid, polyvinylpyrrolidone K30, starch and sucrose, and their Raman spectral data were collected. The components (ICS ) were solved by DSCS-ICA method, and the ICS was compared with the source spectrum, and the correlation coefficient r was used to judge the consistency between ICSand the source spectrum. The results show that, compared with FastICA, the effect of DSCS-ICA is significantly improved, and the correlation coefficient r of ICSand the source spectrum is above 0.99, which shows that the results have good reliability and correspondence. This case can provide reference for the reverse study of drug prescription ingredients, and can be applied to the qualitative analysis of other multi-component systems.