Fuzzy Entropy Based Neuro-Wavelet Identifier-Cum-Quantifier for Discrimination of Gases/Odors

In this paper, a new approach to design of an odor/gas identifier-cum-quantifier is presented. Dynamic response curves of an oxygen-plasma treated thick-film tin oxide sensor array exposed to four different gases were subjected to continuous wavelet transform (CWT). Appropriate wavelet coefficients were selected using multiscale principal component analysis (MSPCA). Fuzzy entropy and fuzzy subsethood values were calculated for the individual odor/gas and for the particular concentration band of each odor/gas, respectively. The quantitative information was encoded in the fuzzy subsethood values of the particular concentration bands in the output feature space, whereas the fuzzy entropy values were used to normalize the training data set consisting of MSPCA selected wavelet coefficients. A feedforward neural network was trained with a backpropagation algorithm with the training data containing the wavelet coefficients normalized with fuzzy entropies of individual odors/gases. The target data set was made up of the fuzzy subsethood values of the particular concentration band. The proposed network achieved identification and quantification of odors/gases with a 100% success rate. Also, fuzzy entropy based normalization helped to achieve 100% identification/quantification with a reduced number of sensors in the array.