Rapid Determination of Vitamin C by NIR Spectroscopy Coupled With CARS-PLS

Abstract

Vitamin C (VC) is a crucial nutritional indicator in fruits, yet its rapid and non-destructive determination remains challenging. This study proposed a competitive adaptive reweighted sampling coupled with partial least squares (CARS-PLS) approach for quantifying VC content in “Jinhuang” mango using near-infrared (NIR) spectroscopy. NIR spectra (1000–2500 nm) of 154 mango samples were acquired, and VC contents were determined by titration. The CARS algorithm selected informative wavelengths, and the CARS-PLS model was compared with full-spectrum PLS and three other wavelength selection methods. Results showed that CARS retained 1078 wavelengths from 1557 variables with a 30.8% reduction while substantially improving predictive accuracy. The CARS-PLS model achieved excellent performance with R²_P of 0.954 and RMSE_P of 0.282 mg/100 g, significantly outperforming full-spectrum PLS with R²_P of 0.884 and RMSE_P of 0.442 mg/100 g. The selected wavelengths were predominantly located in spectral regions associated with carbon-hydrogen (CH) and oxygen-hydrogen (OH) stretching overtones, specifically between 1100 and 1300 nm, 1600 and 1800 nm, and 2200 and 2400 nm. These findings demonstrate that CARS-PLS provides a robust, parsimonious tool for rapid, non-destructive mango VC assessment, with potential applications in quality control and postharvest management.