OPTIMIZATION OF ELECTRONIC NOSE MEASUREMENTS - PART I - METHODOLOGY OF OUTPUT FEATURE-SELECTION

Although very often cited in publications dealing with food products, electronic noses still pose many problems. One is the extraction of fe atures from the response curve; in general, only the adsorption maximu m is retained and input into a classification system. This paper descr ibes a statistic-based methodology developed to extract the most perti nent features from the outputs of SnO2-gas sensor array. Several featu res are extracted from each sensor curve and also from its primary and secondary derivatives. They are then sorted, taking into account thre e specific indexes, designed to describe the repeatability, discrimina tion power and their redundancy. To generalise this approach, the acqu isitions are carried out using various operating conditions. This prot ocol is applied to a set of model mixtures, representing wine with sat isfactory and unsatisfactory tart or vinegar flavour. This paper shows that relevant information can be obtained from the curve maximum, but also from features related to derivatives. Moreover, the most efficie nt features are the same for the five sensors, which would seem to ind icate that they should also be the most suitable ones for all SnO2 sen sors. (C) 1998 Elsevier Science Limited. All rights reserved.