Prediction of chilling injury and key indicator parameters in sweet potatoes using VNIR spectroscopy

Abstract

Chilling injury (CI) in sweet potatoes, which leads to the undesirable accumulation of reducing sugars and renders them unsuitable for processing, is usually assessed through destructive analysis. This study was conducted to investigate the possibility of nondestructive estimation of CI and indicative parameters in intact sweet potatoes. CI indices were developed to estimate CI in ‘Beni haruka’ and ‘Hogammi’ sweet potato cultivars using firmness, total soluble solids (TSS), and malondialdehyde (MDA) as key indicators. Then, visible/near-infrared (VNIR) spectra in interactance mode, along with actual measurements of firmness, TSS, MDA, and CI index, were collected from 160 roots of each cultivar stored at 4 ℃ for 30 days. The data were categorized into calibration, cross-validation, and prediction sets, with preprocessing applied to optimize prediction models. After evaluating the predictive power of raw spectra with spectra preprocessed using various methods such as standard normal variate (SNV), multiplicative scattering correction (MSC), Savitzky-Golay 1st and 2nd order derivatives, partial least square regression (PLSR) was used to develop models correlating VNIR spectra with reference values of TSS, MDA, and the CI index. The models demonstrated promising results, and the regression coefficients for the prediction (R_p²) between the VNIR spectra and the measured values of TSS, MDA, and the CI index were 0.936, 0.902, and 0.812 for ‘Beni haruka’ and 0.847, 0.896, and 0.825 for ‘Hogammi’, respectively. The low standard error of prediction and bias values further indicated the model’s robustness, making these models fast and cost-efficient alternatives to traditional destructive methods for CI analysis. Further research on different cultivars could further improve model accuracy and reliability.