Metodologias analíticas baseadas em imagens digitais e espectroscopia NIR para verificação de adulteração em páprica doce

Abstract

Spices are products made from parts of the same plant species and are used to give flavor, color, aroma and characteristics to foods and beverages. Paprika is a world-renowned spice produced from red peppers. It is sold in small quantities, but has a relatively high value, leading to adulteration practices by traders and producers with the aim of obtaining economic gains. In order to identify these adulterations, it is necessary to develop detection methodologies. In this sense, this work proposes the development of analytical methodologies, based on NIR spectrometry and digital images, for the detection of adulteration in sweet paprika with starch and annatto. For this purpose, the SPA-LDA, GA-LDA and PLS-DA models were created and compared, using NIR spectra with different preprocessing methods and the histograms of digital images generated from the different color channels. In the classification using NIR, for samples adulterated with starch, all models presented satisfactory results, with a correct classification rate of 100%, except for the GA-LDA model using the pre-processed data with Savitzky–Golay second derivative with 11-point window, which achieved a correct classification rate of 94%. The PLS-DA models built for the classification of samples adulterated with annatto classified all samples correctly when the Savitzky–Golay first derivative with 7-point window and Savitzky–Golay second derivative with 11-point window were used. As for the LDA models, the data without pre-processing achieved a 100% correct classification rate using SPA in variable selection, while the GA-LDA models presented better performance with the offset baseline correction and 2nd Der. S.G 11*. In the classification using digital images, when the PLS-DA was built to identify adulteration with starch, the correct classification rate was 0.9400 and with annatto as adulterant it obtained 0.8000. The best result of the SPA-LDA for both adulterants was achieved when the G (green) channel was used, with a correct classification rate of 0.9400 for starch and 0.8200 for annatto. For the GA-LDA model, with starch as adulterant, the RGB + HSI combination presented more satisfactory results, with a correct classification rate of 92%. For the annatto adulterant, the best results were achieved with the G channel, with an 88% correct classification rate. Finally, based on the results obtained in this research, the proposed methodologies present advantages, since they use fast and low-cost techniques, being promising alternatives for the identification of sweet paprika adulterations with starch or annatto.