TDLAS gas concentration retrieval application via SPFE-WV with deep learning

In this study, a deep multilayer perceptron machine network (DMLP) framework that integrates feature engineering and weighted voting integration methods, called spectral profile feature engineering and weighted voting method (SPFE-WV), is proposed. In order to accurately predict the gas concentration in an experimental environment where various noises and background baseline interferences are present, through the preprocessing of the second-harmonic data time series, a feature engineering and weighted voting integration method to successfully extract key features such as the height and area of the main peak and the height difference between different concentrations.