Random Forest Regression for Improving the Measurement Range of a Temperature Interferometric Sensor

In this work, a random forest regression was used to predict the temperature of an interferometric optical sensor over a wide measurement range, overcoming several times the $2\pi $ ambiguities. In particular, in the Fabry-Perot interferometer, this phenomenon is related to the free spectral range (FSR) of the fringes. Additionally, spectral features such as wavelength and amplitude of fringe peaks usually present nonlinear relationships with the target physical variable, in this case temperature. Here, it is shown that by using a random forest (RF) regression it is possible to overcome several FSR ambiguities, to widen the measurement range by a factor of 8 from 5.2 to 49.6 °C, with a RMSE of 0.04554 °C and a MAE of 0.0317 °C.