Power Scaling of Chemiresistive Sensor Array Data for Odor Classification

Abstract The steady state responses of chemical sensors like the tin-oxide and composite conducting polymer sensors exhibit power law dependency on the vapor concentration. In this research, it is shown that linearization of measured (raw) sensor signals by an inverse power scaling improves class separability in feature space, hence the classification efficiency of electronic noses based on these sensors. The follow up data processing is done by dimensional autoscaling, principal component analysis and backpropagation neural network. The data from three tin-oxide sensor array and one CPC sensor array collected from published sources are utilized. It is found that by preprocessing the data according to the suggested power law improves class separability and classification efficiency. An improvement of 14% to 16% is obtained for tin-oxide sensor array and of 33% for conducting polymer sensor array are obtained. The approximate values of the scaling exponents were first deduced as prompted by the sensor response models, and then one adjust were done empirically.