Improved E-nose detection using initial reaction smellprint and advanced classifiers

Abstract

This paper presents a new smellprint derived from Cyra-nose 320 electronic nose, and a robust classification method. The new smellprint is based on the initial reactions of the chemiresistors rather than the bulk relative resistance change. This paper also presents a robust classification method employing Support Vector Machine method. Various combinations of the two smellprints-including their projections to a small number of principal components, are analyzed. The binary Support Vector Machine classification results are filtered through two different mechanisms; a set threshold on the total vote, and a winner-take-all method The classification accuracy is determined through the leave-one-out procedure. The developed system is used for identifying 5 compounds. Promising results are obtained in terms of improved detection at low concentrations and reduced false alarm rates.