Enhanced baseline correction for Raman spectroscopy using a hybrid deep learning approach

Abstract

This research introduces an enhanced baseline correction method for Raman spectroscopy, combining a hybrid deep learning approach with traditional techniques such as polynomial fitting, Gaussian functions, and other nonlinear components. The proposed method significantly improves the signal-to-noise ratio (SNR), achieving up to a tenfold increase over raw spectra and outperforming conventional algorithms such as Imodpoly (polynomial fitting) and AirPLS (Penalised least squares). With a processing time of just 1.07 seconds, the method is well-suited for real-time applications in portable Raman spectroscopy systems. This improvement is critical in Raman spectroscopy, where background noise often obscures weak spectral features, making a high SNR essential for accurate chemical analysis. The rapid processing capability allows for immediate correction of spectral data, ensuring efficient and accurate analysis in practical applications. Thus, this hybrid approach establishes itself as a robust and effective solution for real-time Raman spectroscopy.