A performance evaluation of vis/nir hyperspectral imaging to predict curcumin concentration in fresh turmeric rhizomes

Abstract

Hyperspectral image (HSI) analysis has the potential to estimate organic compounds in plants and foods. Curcumin is an important compound used to treat a range of medical conditions. Therefore, a method to rapidly determine rhizomes with high curcumin content on-farm would be of significant advantage for farmers. Curcumin content of rhizomes varies within, and between varieties but current chemical analysis methods are expensive and time consuming. This study compared curcumin in three turmeric (Curcuma longa) varieties and examined the potential for laboratory-based HSI to rapidly predict curcumin using the visible–near infrared (400–1000 nm) spectrum. Hyperspectral images (n = 152) of the fresh rhizome outer-skin and flesh were captured, using three local varieties (yellow, orange, and red). Distribution of curcuminoids and total curcumin was analysed. Partial least squares regression (PLSR) models were developed to predict total curcumin concentrations. Total curcumin and the proportion of three curcuminoids differed significantly among all varieties. Red turmeric had the highest total curcumin concentration (0.83 ± 0.21%) compared with orange (0.37 ± 0.12%) and yellow (0.02 ± 0.02%). PLSR models predicted curcumin using raw spectra of rhizome flesh and pooled data for all three varieties (R2 c = 0.83, R2p = 0.55, ratio of prediction to deviation (RPD) = 1.51) and was slightly improved by using images of a single variety (orange) only (R2c = 0.85, R2p = 0.62, RPD = 1.65). However, prediction of curcumin using outer-skin of rhizomes was poor (R2 c = 0.64, R2p = 0.37, RPD = 1.28). These models can discriminate between ‘low’ and ‘high’ values and so may be adapted into a two-level grading system. HSI has the potential to help identify turmeric rhizomes with high curcumin concentrations and allow for more efficient refinement into curcumin for medicinal purposes.