Utilização de imagens digitais e calibração multivariada para a determinação da porcentagem de cacau em chocolates

Abstract

Chocolate is one of the most consumed foods in the world, and its production is diverse, presenting products with different cocoa contents. Therefore, this study aimed to develop and validate a calibration model for determining the percentage of cocoa in chocolates by integrating a digital image and a chemometric tool of partial least squares (PLS) regression. A total of 1701 images were acquired, using three different smartphones in a controlled lighting environment. The varieties of milk, semi-sweet and bitter chocolate, from different brands and batches, were considered. As a reference method, the cocoa percentage was obtained from the information on the packaging. The R,G,B histograms were separated into a calibration and external validation set. The choice of the appropriate number of latent variables was made by cross-validation in continuous blocks of 10 samples. The model was built with 10 latent variables, with mean center preprocess. The parameters of merit such as accuracy, adjust, linearity, the inverse of analytical sensitivity, limits of detection and quantification, and the relation prediction deviation showed that the model is suitable for determining the percentage of cocoa in chocolate samples. The regression coefficients allowed to identify the contribution of each channel in the PLS model to predict the percentage of cocoa. The proposed method stands out for its efficiency, practicality, and potential for application in the food industry. Their approach, combined with validation results, suggests a significant contribution to the analysis of chocolates. This research not only presents a solution for determining the percentage of cocoa in chocolates but also paves the way for future investigations and applications in the field of food analysis through advanced imaging and multivariate modeling techniques.