Machine Learning for Optical Gas Sensing: A Leaky-Mode Humidity Sensor as Example

Abstract

Optical gas sensing attracts growing attention in the recent years. This is governed by progressive availability of optical nanostructures fabrication and complex techniques of optical spectrum processing. In the present paper, a room-temperature optical humidity sensor based on a hydrophilic polymer Nafion is theoretically and experimentally investigated. Sensor geometry is optimized for maximum sensitivity of an angle-resolved ATR dip in the Kretschmann configuration. The reflectance dip is attributed to the 2nd order Nafion layer leaky waveguide mode hybridized with the surface plasmon-polariton at the silver/Nafion interface. Results of the relative humidity (RH) retrieval with the regression of the physical model parameters is compared to these obtained with different machine learning (ML) techniques. It is shown that a limited raw data set is enough for using ML algorithms. Accuracy of 0.3% has been demonstrated in RH measurements.