A novel infrared spectral preprocessing method based on self-deconvolution and differentiation in the frequency domain

Spectral preprocessing is a crucial step before establishing a multivariate calibration model for Fourier transform infrared (FTIR) spectroscopic analysis. However, a single preprocessing method has limited effects, such as insufficient noise reduction or limited resolution improvement. As two classical preprocessing methods, Fourier self-deconvolution (FSD) and Fourier filtering and differentiation (FFD) have their benefits and drawbacks. FSD has low noise but limited resolution, while FFD could improve resolution and eliminate baseline simultaneously, but is susceptive to noise. By combining FSD and FFD, we design a novelty resolution enhancement method, termed Fourier self-deconvolution differentiation (FSDD), which can significantly improve spectral resolution with good noise and baseline robustness. Partial least squares regression (PLS) is combined with five preprocessing methods, FSD, FFD, direct derivation (DD), Savitsky-Golay smoothing and derivation (SGSD), and FSDD, to analyze the albiflorin dataset and the γ-polyglutamic acid (γ-PGA) fermentation dataset. Compared with other preprocessing methods, FSDD obtained the optimum performance merits, including root mean squared error of prediction (RMSEP), determination coefficient (R2), and Mean Absolute Error (MAE).