Establish a model for a portable near-infrared spectroscopy (NIR) device in determining total carbohydrate content in foods

Abstract

This study was to develop a calibration model for a portable near-infrared (NIR) spectroscopy device to determine the total carbohydrate content in processed foods, especially for meat-free products. NIR spectra in the 740 - 1070 nm region of 251 food products were analyzed. The results provided new insights into the shortwavelength NIR spectroscopy. The model between the total carbohydrate content measured by traditional methods and the spectral data collected from the NIR device was built with the partial least square (PLS) algorithm combined with the crossvalidation technique by the SCiO lab software (Consumer Physics, USA). The effects of number of latent variables (LVs) on the model fitting was also investigated. The results showed that with LVs = 30 a high correlation coefficient R2 of 0.956 and a low root mean squared error (RMSE) of 3.88 were obtained, indicating the good fitting of the model. The correlation coefficient R2 between the predicted and measured values of 62 testing samples used for model validation was 0.860 with the standard error of prediction (SEP) of 4.15, re-verifying the high accuracy of the model. The performance values achieved in this study indicated the possibility of the NIR method to become a popular one for the fast determination of total carbohydrate content in foods is fully promising. Key words: Near-infrared spectroscopy, NIR, SCiO, partial least square, total carbohydrate content, calibration