Improving Quantitative Analysis with Cross Instrument-Sparse Bayesian Learning (CI-SBL) Raman Spectroscopy Analysis Algorithm.

Abstract

Qualitative and quantitative analysis of Raman spectroscopy is a widely used nondestructive analytical technique in many fields. It utilizes the Raman scattering effect of lasers to obtain molecular vibration information on samples. By comparison with the Raman spectra of standard substances, qualitative and quantitative analyses can be achieved on unknown samples. However, current Raman spectroscopy analysis algorithms still have many drawbacks. They struggled to handle quantitative analysis between different instruments. Their prediction accuracy for concentration is generally low, with poor robustness. Therefore, this study addresses these deficiencies by designing the cross instrument-sparse Bayesian learning (CI-SBL) Raman spectroscopy analysis algorithm. CI-SBL can facilitate spectroscopic analysis between different instruments through the cross instrument module. CI-SBL converts data from portable instruments into data from scientific instruments, with high similarity between the converted spectrum and the spectrum from the scientific instruments reaching 98.6%. The similarity between the raw portable instrument spectrum and the scientific instrument spectrum is often lower than 90%. The cross instrument effect of the CI-SBL is remarkable. Moreover, CI-SBL employs sparse Bayesian learning (SBL) as the core module for analysis. Through multiple iterations, the SBL algorithm effectively identified various components within mixtures. In experiments, CI-SBL can achieve a qualitative accuracy of 100% for the majority of binary and multicomponent mixtures. On the other hand, the previous Raman spectroscopy analysis algorithms predominantly yield a qualitative accuracy below 80% for the same data. Additionally, CI-SBL incorporates a quantitative module to calculate the concentration of each component within the mixed samples. In the experiment, the quantification error for all substances was below 3%, with the majority of the substances exhibiting an error of approximately 1%. These experimental results illustrate that CI-SBL significantly enhances the accuracy of qualitative judgment of mixture spectra and the prediction of mixture concentrations compared with previous Raman spectroscopy analysis algorithms. Furthermore, the cross instrument module of CI-SBL allows for a flexible handling of data acquired from different instruments.