The analysis of sensor array data with various pattern recognition techniques

A sensor array composed of selective and partially selective electrodes is applied to milk recognition. The task of the system is to distinguish among five brands of milk. For this purpose, five pattern recognition (PARC) procedures are employed: three linear (K-nearest neighbours, partial least squares, soft independent modelling of class analogy) and two nonlinear (back propagation neural networks and learning vector quantization). Classification accuracy is compared and some analogies with general rules referring to electronic nose were found. LVQ networks are proved to exhibit the best performance. Their further advantages, such as fast training and robustness, make them the suggested pattern classifiers for sensor array data.