Discrimination and sensory description of beers through data fusion

Abstract

Beer samples of the same brand and commercialized as a same product, but brewed in four different factories were analyzed with three techniques, an MS e-nose, a mid-IR optical-tongue and a UV–visible, to see if the factories show differences and to find out if the differences found could be attributed to different sensory properties. The data from the three instruments were fused to improve the ability of classification with respect to the individual use of the techniques. Two levels of data fusion were studied: low and mid level fusion, and the classification was performed by linear discriminant analysis (LDA). Mid-level fusion provided better classification results (above 95% correct classification) than those of low-level fusion and also than those obtained when using the individual techniques. Moreover, by means of the score and loading plots obtained by Fisher-LDA, it was possible to interpret the chemical information provided by the three techniques, and we were able to relate the variables associated to each sensor to the main compounds responsible of the sensory perception.