Improving the training and data processing of an electronic olfactory system for the classification of virgin olive oil into quality categories

A large amount of virgin olive oil samples and different feature selections have been employed in order to improve the classification capacity of an electronic olfactory system, based on thin film metal oxide semiconductor, to be applied into the virgin olive oil industrial analysis, according to the current European Union regulation. The basic ability of the employed equipment for aroma discrimination was tested by using pure aromatic compounds some of them usually present in the virgin olive oil. For these samples, the employed system shows well different responses and identification patterns. After that, more than three hundred virgin olive oil samples from different origins were analysed. The fingerprints of such samples anticipate good discrimination between higher and lower quality samples. Multivariate analysis (principal component analysis and discriminate factorial analysis) was used in order to obtain the best pattern recognition algorithms and the better classification of unknown samples. The real improvement was obtained when an alternative feature selection named Five Contiguous Points was used. This feature selection was revealed to increase the separation between the different quality categories due to a better use of the information contained in the sensor response curves. In summary, the best results for virgin olive oil identification were obtained by the Five Contiguous Points–discriminate factorial analysis combination, with the increase of the number of samples demonstrating that the classification ability is clearly improved. Finally, this study also provides an overview on future application.