vis-nir hyperspectral imaging coupled with chemometrics for turmeric authentication

Ensuring food quality is a vital issue because food plays an essential role in human health [1]. spices are precious and are often ground or powdered, which can cause many counterfeits by using additives and different colors [2]. one of the most popular spices in the world is turmeric (curcuma longa), which is used in various fields such as food and medicine [3]. in the present contribution, the visible-near infrared hyperspectral imaging (vis-nir-hsi) system combined with different chemometric methods is proposed as a novel technique for authentication and detecting five common adulterants (corn flour, rice flour, starch, wheat flour, and zedoary) in turmeric powder. the vis-nir hyperspectral images of three datasets were recorded and arranged in different datasets including 23 authentic samples (dataset i), binary mixtures of pooled turmeric and each adulterant in seven concentration levels (1-35%) (dataset ii) and mixtures of five plant materials and turmeric in different percentages by d-optimal design (dataset iii). then, multivariate curve resolution-alternating least squares (mcr-als) and mean-field independent component analysis (mf-ica) were used as multivariate resolution methods to exploit pure spatial (distribution map) and spectral profiles of the pure components [4]. due to the presence of rotational ambiguities, the performance of these two algorithms were compared by considering the area of feasible solutions (afss) using fackpack. the resolved spatial profiles (distribution maps) of turmeric was then used to find patterns of authentic samples by using principal component analysis (pca). in the next step, data-driven soft independent modeling of class analogy (dd-simca) was used for modelling of the distribution maps obtained by mcr-als and mf-ica. on this matter, the model sensitivity and specificities obtained based on mcr-als resolved distribution maps of authentic turmeric samples were better than those obtained by mf-ica. finally, partial least squares-discriminant analysis (pls-da) was utilized for supervised classification and distinguishing between authentic and adulterated turmeric samples again by using distribution maps obtained by mcr-als and mf-ica. again good model accuracies were obtained for calibration and prediction sets (100% for mcr-als and 96% for mf-ica). to test the applicability of the proposed method in mixed samples, dataset iii was analyzed by the developed dd-simca and pls-da models and good classification results were obtained. this was proof that the proposing model can detect adulterants in turmeric powder even at one percent of the adulteration level. therefore, it is concluded that vis-nir hsi combined with chemometric methods is a powerful and novel technique for turmeric authentication and adulteration detection.