Data fusion in electronic tongue for qualitative analysis of beers

This paper presents the development of an Electronic Tongue based on two different arrays of electrochemical sensors (i.e. potentiometric and voltammetric) for the identification of three styles of beer. Conventionally, electrochemical measurements contain hundreds of records and cannot be processed directly, due to its high data dimension. Therefore, information obtained from both sensor families was prepossessed in order to extract representative features and then fused to improve the classification ability regarding to the use of single sensor data. On the one hand, Discrete Wavelet Transform and statistical procedures were employed as feature extraction techniques. On the other hand, classification model was build using Linear Discriminant Analysis and validated by Leave-one-out cross-validation procedure. Final results demonstrate that the ET employing data fusion is able to distinguish 100% of the types of beer as well as its manufacturing process.