Língua Eletrónica: uma ferramenta prática na análise sensorial de azeitonas de mesa

Abstract

Table olives are consumed and appreciated all around the world and although its commercial classification is not legally required, the International Olive Council suggests the implementation of specific regulation based on the organoleptic evaluation carried out by sesnory panel. The implementation of a trained sensory panel requires the compliance with the guidelines established by the International Olive Council, being a complex and time-consuming task, which evalautions are not without subjectivity. In this work, first, an electronic tongue was used to classify table olives according to commercial categories established based on the presence and median intensities of the predominantl organoleptic defects perceived by the taste panel. Linear discrimination models were established based on the most informative subsets of potentiometric signals recorded by the electronic tongue sensors and selected using the simulated annealing algorithm. The qualitative predictive performance of the established classfication models were evaluated using the leave-one-out cross-validation and the repeated K-folds cross-validation techniques, the latter the minimizes the risk of overfitting, allowing more realistic results. The potential of this qualitative approach based on electrochemical profiles generated by the electronic tongue, has been satisfactorily demonstrated by: (i) the correct classification (sensitivity ≥ 93%) of standard solutions (n-butyric acid, 2-mercaptoethanol and cyclohexanecarboxylic acid) according to the sensory defects that they mimicked (butyric, putrid or zapateria); (ii) the correct classification (sensitivity ≥ 93%) of reference samples of olives and brines solutions (possessing a single intense defect) according to the type of default perceived by the panel (winey-vinegary, butyric, musty, putrid or zapateria ) and selected by the taste panel; and (iii) the correct classification (sensitivity ≥ 86%) of table olives samples despite their great heterogeneity, which presented one or more organoleptics defects, identified by the sensory panel during the olive and/or brine solutions evaluation, according to their commercial category (extra olive without defects, extra, 1st choice, 2nd choice and olives that can not be sold as table olives). Finally, the electronic tongue ability to quantify the median intensities of the negative attributes perceived by the sensory panel in table olives was demonstrated using multiple linear regression models based on subsets of potentiometric signals generated by the electronic tongue during the table olives and brine solutions analysis, and selected using the simulated annealing algorithm. The predictive performances of the quantitative models was evalauted using the same two cross-validation techniques. The established model, for each of the five sensory defects perceived in the table olives samples allowed to satisfactorily quantify the medians of defects intensities (R² ≥ 0.97). Thus, based on the satisfactory quality and quantitative results achieved allows to foresee, for the first time a possible practical application of electronic tongues as a sensory tool for table olives’ defects assessment, being a fast, economic and useful technique for the organoleptic evaluation of negative attributes, complementary to the traditional sensory evaluation carried out by trained sensory panels.