Using the convolutional neural network for accurate gas discrimination and concentration measurement by an ultrasonically radiated thermoelectric gas sensor

Ultrasonic radiation can catalyze the redox reaction at the sensing surface of a gas sensor, which provides a new method to significantly raise the gas sensors’ sensitivity and to discriminate gas species. In this work, a convolutional neural network (CNN) based model for species discrimination and concentration regression is proposed and developed for an ultrasonically assisted thermoelectric gas sensor, in order to achieve accurate results. The experiment evaluates the gas sensor’s ability to detect and quantify four gases in clean air: acetone, methanol, hydrogen and ethanol. The sensor achieves a high accuracy of 98.75% for gas identification and a low error of 4.5% for gas concentration measurement, respectively, within the 2%–20% lower explosive limit (LEL) range.