Near-infrared spectral quantitative analysis with reduced moisture interference using a spectral decomposition optimization algorithm: A case study on Amadori compounds in tobacco leaves

The strong absorption characteristics of water in the near-infrared (NIR) spectral region significantly hinder the quantitative analysis of chemical compounds. Using the quantitative analysis of Amadori compounds in tobacco leaves as a case study, this study proposed a spectral decomposition optimization algorithm (SDOA) to mitigate water interference in NIR spectra, which helped improved the accuracy of predictive models. The method utilized a reconstructed spectral matrix derived from aged tobacco leaves across 13 moisture levels, ranging from 4 % to 16 % water content. By constructing a difference spectrum matrix, performing singular value decomposition, and applying projection transformations to the original NIR spectra, the influence of water on the spectral was reduced. A partial least squares (PLS) model was established based on the moisture-corrected spectra to predict the content of 17 Amadori compounds in tobacco. The results demonstrated that: (1) After SDOA processing, the average correlation coefficient ( R ‾ ′ 2 ) of spectra at different moisture levels increased from 0.9903 to >0.9999, indicating significantly enhanced spectral consistency; (2) Among the 17 Amadori compounds, Fru-Pro, which was present at a relatively high concentration in tobacco leaves, exhibited minimal water interference. For the remaining 16 compounds, the SDOA-PLS models achieved robust predictive performance, as evidenced by evaluation metrics including the coefficient of determination ( R 2 ), root mean square error (which ranged from 0.8705 to 0.9859), and relative prediction deviation (which ranged from 2.778 to 8.423). (3) One-way analysis of variance confirmed no significant differences (α = 0.05) in the predicted Amadori compound contents across tobacco samples with varying moisture levels using the SDOA-PLS method. The SDOA-PLS model effectively eliminates water interference in NIR spectra and significantly improves the accuracy and stability of Amadori compound content prediction in tobacco leaves. • The SDOA reduces moisture interference on near-infrared spectroscopy. • The SDOA-PLS model reliably predicts 16 Amadori compounds in tobacco. • The SDOA-PLS model achieves moisture-independent predictions across 4 %–16 % moisture contents.