Building a calibration model for a portable NIR device in quantitative analysis of the protein content of meat analogues and vegetarian food

Abstract

In recent years, the invention of the portable low-cost near-infrared (NIR) devices has enabled even consumers to know the nutrition of what they eat anywhere and anytime. The objective of this research was to investigate the possibility of using NIR in determining protein content in common meat analogues. Totally 246 calibration samples and 54 validation samples were used. The spectral data of calibration samples were preprocessed using 1st derivative, followed by Standard Normal Variate (SNV) before modelling. Cross-validation technique was used and followed by the algorithm Partial Least-Squares Regression (PLSR) for the development of the calibration models. The results exhibited that the optimal predictive model was fairly good with a high correlation coefficient R2 value of 0.884 and a low root mean square error (RSME) value of 5.03. The results of model validation using 54 samples showed a high R2 value of 0.85 and a low RSME value of 5.74. The error was mostly contributed by the standard error while the bias error was negligible. This suggests the potential of NIR devices in practical application of quick chemical analyses. Key words: NIR; SCiO; protein content; processed food; meat analogues; PLSR