New pattern recognition systems designed for electronic noses

Electronic noses represent a big challenge for the pattern recognition (PARC) systems due to several particular problems they involve. The work presented in this paper is targeted to develop specific methods for these kinds of problems. One of the main issues to deal with, is the concentration variation, as a main cause of pattern dispersion in aroma/gas recognition. Such dispersion hinders easy cluster separation, specially for small aroma intensities. Specific algorithms for gas identification are introduced. They cope with the usual elongated cluster structure found in electronic noses. The PARC systems combine self-organising maps (SOM) and minimum spanning tree (MST) to build curvilinear prototypes. The method is exemplified with a minimal tin dioxide sensor array chosen for CO and CH4 detection in domestic premises.