High performance of gas identification by wavelet transform-based fast feature extraction from temperature modulated semiconductor gas sensors

Abstract

Temperature modulation of semiconductor gas sensors is a powerful strategy to improve selectivity and stability of gas sensors in applications of identifying different gases. This paper presents a new strategy to extract features from the response of a thermally modulated semiconductor sensor for gas identification. This strategy contains two main steps. First, make the sensor work under temperature modulation and pre-process the sensor's response signal. The presented pre-processing method can suppress the influence of environmental temperature or humidity on sensor response. Secondly, apply wavelet decomposition to extract features of the pre-processed sensor's response curves. The rules of selecting the resulted wavelet coefficients as the characteristic variables for gas identification are also presented. Experiments results show that the strategy proposed here allows accurate discrimination between gases studied (hydrogen, carbon monoxide and their binary mixture) over a wide concentration range, from 10 to 1000 ppm, by using only one commercial metal oxide gas sensor and owns good ability of restraining the influence of gas concentration variation, sensor's cross sensitivity and sensor drift on the identification result. Experiments also show that the proposed strategy outperforms FFT for the featured extraction from the response of thermally modulated semiconductor gas sensor.