Evaluation of a miniaturized NIR spectrometer for estimating total curcuminoids in powdered turmeric samples

Abstract

Commercial availability of miniaturized spectrometers, equipped with machine learning, and cloud computing capabilities, is transforming the food-testing industry by enabling instant results and on-the-spot decision making. To demonstrate this, we have evaluated SCiO(TM) from Consumer Physics to quantify curcumin in turmeric with reflectance spectroscopy in the NIR region (740-1050nm). Different pre-processing combinations were tried to maximally extract useful information. The decision of the best combination was based on models built with Partial Least Squared Regression (PLSR) algorithm. The best combination yielded a model with a coefficient of determination ($R2$) of 0.797 and a root-mean-squared error (RMSE) of 0.306. This was validated on a test set of 6 samples and gave a high $R2$ of 0.93. This study shows potential for similar, instant quality analysis of other powdered spices with commercial spectrometers and optimized machine learning models.