Transient feature extraction for machine olfaction based on Wavelet decomposition

The performance of electronic olfaction devices is highly dependent on the quality of input characteristics obtained from sensorspsila response. These units collect information of the odors they are assessing using an array of gas sensors. Typically, these devices have a high-dimensional inpsut space which makes odor classification difficult and requires time-consuming computation. In this study, and requires time-consuming computation. In this study, a multiresolutional approximation technique from the Discrete Wavelet Transform (DWT) is employed to capture only relevant features of the sensor arraypsilas dynamic responses. Three families of wavelets are evaluated using three statistical and neural network classifiers (k-nearest neighbor, Backpropagation, and RBF neural networks) for two differential odor data sets (coffee and soda). The classification experimental results show promising improvements when compared to conventional steady-state classification performance.