Baseline-Free Wide-Range Methane Concentration Measurement Based on Logarithmic Second Derivative Spectroscopy

Abstract Methane (CH4) leak detection requires not only high detection sensitivity but also accurate concentration measurements over a wide concentration range. To address the difficulty of baseline extraction in tunable diode laser absorption spectroscopy based on direct absorption spectroscopy (TDLAS-DAS) and detector signal saturation under high absorption conditions, a logarithmic second derivative convolutional neural network (LSD-CNN) is proposed for baseline-free CH4 concentration measurement over a wide concentration range. By applying a logarithmic transformation followed by second derivative processing to the transmitted spectrum, a new spectral feature domain is constructed. This transformation removes the influence of the baseline while preserving the relationship between the transformed spectrum and CH4 concentration. Considering that the spectral characteristics in the feature space differ significantly between unsaturated and saturated conditions, global standardization and pointwise standardization are used to construct a dual input, which is then processed by a one dimensional CNN for CH4 concentration inversion over a wide concentration range.The proposed method is validated using a mid-infrared CH4 sensor. Experimental results show that the coefficient of determination (R2) reaches 0.99991 over the concentration range from 500 ppb to 10000 ppm. Allan-Werle deviation analysis gives a minimum detection limit of 2.63 ppb at an integration time of 98 s.