Robust gas quantification in mid-infrared FTIR spectroscopy via a suppression-adaptation-optimization model

Mid-infrared Fourier transform infrared (FTIR) spectroscopy is a widely adopted technique for multi-gas detection due to its high sensitivity and capability for simultaneous quantification. However, instrumental noise and environmental disturbances (e.g., ambient light interference and temperature variations) introduce spectral fluctuations and interferences, making accurate FTIR quantification particularly challenging. In response to these issues, we propose a suppression-adaptation-optimization (SAO) model that builds upon the physics-based forward model, using it as the foundational reference for residual correction and iterative optimization to improve quantification robustness under practical measurement conditions. The SAO model consists of three stages: (1) noise suppression using linear or nonlinear filtering to enhance signal quality; (2) residual adaptation, where a generalized loss function is adapted to penalize the residual between the denoised spectra and the spectra simulated by the physics-based forward model, weakens the reliance on the independent Gaussian assumption; and (3) loss function optimization, performed using the Yogi optimizer, which iteratively updates the model parameters to minimize the average loss across all data points. As a result, the quantitative impact of spectral deviations can be effectively mitigated through denoising and residual adaptation. We evaluated the SAO model using both simulated and experimental transmission spectra with RMS noise approximately 1×10^-3 in the spectral range of 2150 cm^-1 to 2310 cm^-1, targeting the retrieval of CO₂, N2O, and CO concentrations. Compared to the Levenberg-Marquardt (LM) method, the SAO model reduces the standard deviation of retrieved concentrations by at least 15 % in simulations and up to 20 % in experimental measurements. It demonstrates that the integration of noise suppression and residual correction enhances the robustness of FTIR gas quantification, revealing the potential of the SAO model in industrial monitoring applications.