E-nose system by using neural networks

This paper considers a new construction of an electronic nose (E-nose) system based on a neural network. The neural network used here is a competitive neural network by the learning vector quantisation (LVQ). Various odors are measured with an array of many metal oxide gas sensors. After reducing noises from the odor data which are measured under different concentrations, we take the maximum values among the time series data of odors. As they are affected by concentration levels, we use a normalization method to reduce the fluctuation of the data and reorder the measurement data according to the concentration levels to make the features invariant with the concentration levels. Those data are used to classify the various odors of teas and coffees. The classification results are about 96% in case of four kinds of teas and about 89% for five kinds of coffees.