Improving pattern recognition of electronic nose data with time-delay neural networks

An enhanced time-delay neural network (TDNN), using time series sensor response data, improved pattern recognition ability of an electronic nose (e-nose) in discriminating four different spices. TDNN was used for analysis of e-nose time series sensor data from 0 to 4 min, while two popular pattern recognition methods, discriminant function analysis (DFA) and multilayer perceptron (MLP) trained by back-propagation, were used to analyze the "instantaneous" e-nose data obtained from sensor responses only at 4 min. The results showed that DFA and MLP correctly recognized 59.4 and 62.5% of the spices respectively while TDNN correctly recognized all (100%) spices. (C) 2003 Elsevier B.V. All rights reserved.