Abstract

Lycopene liquid beverages are functional products with strong antioxidant activity, but their quality control is hindered by the lack of rapid and simultaneous detection methods for multiple indicators. This study investigated a near-infrared spectroscopy (NIRS)—based approach for the simultaneous quantification of pH, moisture, and lycopene content in lycopene liquid beverages. A combined preprocessing strategy integrating Savitzky-Golay smoothing and standard normal variate (SG-SNV) was optimized via grid search to reduce spectral noise and scattering effects. Model interpretability was further enhanced by SHapley Additive exPlanations (SHAP) analysis. Furthermore, four variable selection algorithms (MWPLS, UVE-SPA, ICO, and CARS) were compared to improve model performance and interpretability. The results showed that SG-SNV preprocessing notably improved the prediction performance for pH (R2CV = 0.9398, R2PRE = 0.9624) and lycopene (R2CV = 0.9860, R2PRE = 0.9576), while moisture prediction remained at a comparable level (R2cv = 0.9617, R2PRE = 0.9727). SHAP analysis identified spectral regions that may be associated with each component, providing chemically plausible explanations. For variable selection, the optimal algorithm differed across the three quality indicators. For moisture, CARS achieved the highest prediction performance (R2PRE = 0.9946), while ICO provided comparable accuracy with slightly better model stability. For pH, UVE–SPA produced the strongest generalization ability (R2PRE = 0.9705) with only 5 selected variables. For lycopene, none of the variable selection methods improved upon the preprocessing-only model (R2PRE = 0.9576); MWPLS retained a parsimonious 21-variable model but with reduced predictive accuracy (R2PRE = 0.8888). This study explores a preliminary rapid and non-destructive NIR analytical approach for the simultaneous detection of multiple quality indicators in lycopene liquid beverages, suggesting the feasibility of chemometric modeling for this application. The results provide a preliminary foundation for future development of rapid multi-indicator analytical methods. However, this work was conducted entirely under controlled laboratory conditions using a small, gradient-designed sample set with no production-line or online validation; it should therefore be regarded as a proof-of-concept feasibility study. Substantial further work—including validation with larger, naturally variable sample cohorts—would be needed before the approach could be considered for practical deployment.

IPC Classification

Keywords