Baseline correction for Raman spectra using penalized spline smoothing based on vector transformation

Abstract

Baseline drift always negatively affects the qualitative or quantitative analytical results of Raman spectroscopy for many types of Raman spectrometers. Several baseline correction methods have been applied for processing Raman spectra. The parameters of these methods, however, are usually set through a complex and time-consuming process. More importantly, existing methods cannot normally provide a good estimation of the dramatically changing baselines. In this paper, we propose a penalized spline smoothing method based on vector transformation (VTPspline) for baseline estimation. The proposed baseline correction method is initiated by the raw spectrum baseline which is set as the original Raman spectrum. Meanwhile a vector v, the number of elements of which is equal to that of the spectral wavenumbers is randomly generated, and all elements are set as 1 to indicate that the corresponding Raman spectral wavenumbers are considered to be background points. Vector transformation is used to transform vector v into a new sequence to make random spectral wavenumbers turn into the suspected characteristic peak channels. Based on the penalized spline smoothing algorithm and iterative process, the points that belong to the spectral background region are automatically and gradually preserved. The performance tests using both simulated and experimental Raman spectral data demonstrate that the proposed method outperforms the other existing methods for baseline estimation, such as adaptive iteratively reweighted penalized least squares and improved asymmetric least squares methods. The results also indicate that the proposed VTPspline method can handle complex and severely drifting baselines well, while maintaining the original characteristic Raman features.