Baseline correction algorithm for Raman spectra based on median filtering and un-uniform B-spline

As one of the key steps for data preprocessing of Raman spectra, baseline correction is an effective method to eliminate fluorescence interference. Traditional algorithms such as polynomial fitting and uniform B-spline fitting are simple and easy to implement, but the uncertain fitting order and internal knots limit their flexibility. In addition, the baseline correction results of traditional algorithms often occur over and under fitting phenomena. Therefore, we propose a baseline correction algorithm for Raman spectra based on median filtering and un-uniform B-spline. Firstly, the trough points of the spectral data are filtered by smoothing preprocess, difference calculation and threshold setting, and the internal knots of the un-uniform B-spline are adaptively selected by the trough positions of the spectral data. Then, the median filtering algorithm is used to process the spectral data so that the un-uniform B-spline has a better baseline fitting effect at the position where the signal changes from peak to smooth band. Finally, the un-uniform B-splines is used to fit the baseline by fitting the baseline iteratively. The proposed algorithm overcomes the shortcoming of traditional B-spline algorithm that the internal knots need to be selected manually based on different Raman spectra, and also avoids influencing the baseline fitting by random noise in the spectral data, and thus further improving the spectral baseline correction effect. The original Raman spectra of polymethyl methacrylate and normal octane are used for experimentally evaluating the baseline correction effect. Compared with the results from polynomial fitting, uniform B-spline and adaptive iteratively reweighted penalized least squares algorithms, the experimental results show that the proposed algorithm can well eliminate the Raman signal baseline drift effectively without over or under fitting phenomena, and it can perform better baseline correction for different baseline drift situations. Therefore, the proposed algorithm can provide more accurate and reliable information for the further analysis of spectral data.