Classification of chocolate according to its cocoa percentage by using Terahertz time-domain spectroscopy

Abstract

Feasibility of a non-destructive classification of chocolate based on its cocoa content was examined by using a Terahertz time-domain spectroscopy system combined with a multivariate analysis. For this purpose, the spectra from 0.5 THz to 10 THz of 5 chocolate samples (50%, 60%, 70%, 80% and 90% of cocoa) were examined. The acquired data matrices were analyzed by using a Fourier Transform, obtaining the dielectric function and the absorbance curve. Based on the latter, samples were classified by using 24 models of mathematical classification, achieving differences of around 93% through the model of Gaussian SVM algorithm with a kernel scale of 0.35 and a one-against-one multiclass method. This was reduced by using a Main Component Analysis, obtaining most of the spectral variations with PC1 (63.8%) and PC2 (36.2%). It was concluded that the combined processing and classification of images obtained from Terahertz time-domain spectroscopy, as well as the use of machine learning algorithms, can be used to successfully classify chocolates with different percentages of cocoa.