Raman spectroscopy is a non-destructive testing technology. Qualitative and quantitative information of substances can be obtained by Raman spectroscopy based on its fingerprint spectrum characteristics and good performance, so it has been widely used in environmental, chemical, biomedical, and other fields. However, in Raman spectrum measurement, the spectral baseline drift caused by fluorescence scattering signal will make it challenging to identify the characteristic peaks of the Raman spectrum, which will have different degrees of adverse effects on the results of qualitative or quantitative analysis. Various baseline correction methods have been successfully applied to multiple spectral processing. However, the parameters of these methods still need to be set according to the experience of the researchers. In this paper, a baseline correction method based on cubic spline smoothing combined with discrete state transformation algorithm (DSTAspline) is proposed, which can find the best estimation result of fluorescence background globally based on discrete state transformation strategy and obtain the optimal baseline curve based on smooth spline curve fitting, thus completing the baseline fitting of Raman spectra signals. In the experiment, the spectra of two kinds of actual substances are selected, and the baseline correction of the Raman spectra is carried out based on the proposed algorithm. The experimental results show that the proposed method can effectively eliminate the baseline drift of Raman spectra with different intensity fluorescence backgrounds, thereby obtaining the real Raman spectra signals without complex parameter setting and adjustment processes. It provides reliable information for further Raman spectra data to realize the qualitative and quantitative analysis of substances.

In our proposed method, cubic spline smoothing is used to fit the baseline drift for Raman spectra. A binary vector V with the same dimension as the number of wavenumber points in the Raman spectrum is initially generated by the algorithm. The element in vector V is coded as either 1 or 0 to represent the corresponding point in the Raman spectrum whether it belongs to the fluorescence background region. Then discrete state transformation algorithm is used to generate new values of the vector V. The cubic spline smoothing method fits the estimated spectral baseline according to the wavenumber points in the fluorescence background region. The root mean square error (RMSE) between the fitted spectral baseline and the number of estimated background points is used to terminate the iterative process. Finally, once the iterative process is terminated, the accurate spectral baseline is fitted with the optimal vector V.

The performance of the proposed DSTAspline baseline correction method was tested and verified by simulated and real Raman spectra (Fig. 1). The airPLS, IAsLS, ASWF and DSTAspline baseline correction methods were applied to the simulated and real Raman spectra for comparing their performances. For the simulated Raman spectrum, it is evident that the baseline drifts in the simulated Raman spectra (without noise or with strong noise) were well estimated by our proposed DSTAspline method, and real Raman signal could be extracted from the original Raman spectrum [Fig. 2(a) and Fig. 3(a)]. More importantly, the RMSE of the proposed DSTAspline method was the smallest among these four methods, even for the simulated Raman spectrum with intense noise [Fig. 2(b), Table 1 and Fig. 3(b)]. For real Raman spectra of the rhodamine B and tricyclazole, the proposed DSTAspline method can precisely estimate the severely drifting baselines and obtain the best performance among the four baseline correction methods (Fig. 4 and Fig. 5). Finally, the proposed DSTAspline method was applied to detecting and quantifying the rapeseed oil adulterated in peanut oil by combining Raman spectroscopy and chemometrics (Fig. 6). The result shows that the proposed DSTAspline method can improve the prediction accuracy of the calibration model (Fig. 7).

The proposed DSTAspline method provides an adaptive and effective way for estimating and correcting the baselines with various curvature curve backgrounds and multiple overlapping peaks. Based on the discrete state transformation algorithm, the spectral wavenumbers in the background region were chosen and the cubic spline smoothing algorithm smoothly fitted the estimated background. The experiment results with the simulated spectra demonstrate that the DSTAspline method provides better results of baseline correction and peak intensity estimation than airPLS, IAsLS, and ASWF. The proposed DSTAspline method avoids the complicated and artificial process of choosing numerous parameters for obtaining optimal performance. Moreover, based on the cubic spline smoothing algorithm, the proposed DSTAspline method is not sensitive to noise. The baseline correction results of experimental Raman spectra also show that the DSTAspline method could handle various types of backgrounds in real Raman spectra. Moreover, the proposed DSTAspline method was used to correct the baseline of the peanut oil samples adulterated with rapeseed oil, and the excellent results have proven that the DSTAspline method can improve the prediction accuracy. The DSTAspline algorithm which was applied in the MATLAB platform and the related data have been made publicly available, free to use for anyone at https: ∥github.com/skyworth0103.