ABSTRACT
These include artificial neural networks that take the input patterns generated by the array of conducting polymer sensors [I], which may be trained to associate these patterns with particular classes of volatile chemicals that may be of interest to the user. Such architectures include 2-layer systems trained by conventional back-propagation of error [2] and radial basis function network [3]. Of particular interest to us is not only the identification of odour classes, but also the prediction of odour concentrations, even when the background may be complex. The quantification of odours is very desirable feature in real life and is much more difficult to predict concentration levels of single chemicals or mixture than classification of different chemicals. Most of researchers have tried to classify the odour identification and to predict concentration levels for odours separately using different
algorithms. In this paper, we have investigated the properties of multilayer perceptron (MLP) for odour pattern classification and concentration estimation simultaneously. When the MLP may be has a fast convergence speed with small error and excellent mapping ability for classification, it can be possible to use for classification and concentration prediction of volatile chemicals simultaneously. However, the conventional MLP, which is backpropagation of error based on the gradient method, was difficult to use for odour classification and concentration estimation simultaneously, because it slow to converge and may fall into the local minimum. We adapted the Levenberg-Marquardt (L-M) algorithm [4,5], having advantages both the steepest descent method and the Gauss-Newton method instead of the conventional steepest descent method for the simultaneous classification and concentration estimation of odours. It is confirmed by experimental trails with the electronic odour sensing system, which has been constructed at UMIST by Dr. Krishna C. Persaud.
