Smell classification of wines by the learning vector quantization method

We consider a classification of white wine and red wine by a learning vector quantization method of competitive neural network. First, we measure smell data using metal-oxide semiconductor gas sensors which change smell data into electrical voltages based on oxidation and reduction processes. Two kinds of wines, white wine and red wine, are classified using smell data. Since a smell density of wine is rather thin, we use a bubbling method to make the density level higher. Here, we adopt a mono trap which is a kind of molecular sieves. By this way we obtain smell data of wines of high concentration level. After absorbing process, we take the temperature of a silica tube from a room temperature to 300 degrees Celsius. Using the learning vector quantization method, we classify two kinds of wines. We show that the classification accuracy rate for the white wine is around 97% and that for the red wine is around 83.4%, respectively.