Near-infrared methane sensor with neural network filtering

Abstract

A novel methane sensor based on neural network filter (NNF) assisted direct absorption spectroscopy (DAS) technique was proposed and experimentally demonstrated. The developed detection device adds the benefits of a digital filter based on the neural network, thereby compensating the shortcomings of traditional DAS. We overcame the scarce data problem by using the simulated absorption spectra that are highly consistent with practical experimental conditions to construct and train the NNF. The proposed NNF showed the best performance compared with several widely used filtering algorithms. We performed a detailed assessment of the NNF-improved detection system. The proposed sensor shows more accurate concentration retrieval and better stability in a real-time measurement. The minimum detection limit of 2.93 ppm∙m (1σ) was obtained, which is a significant improvement compared to previous reports of near-infrared methane detection with the DAS technique. Finally, we systematically discuss the frequency principle underlying the NNF to explicitly interpret the mechanism of the generalized filtering. The improved methane sensor proves the feasibility of enhancing the performance of DAS technique with the neural network algorithm and broad applicability of this approach to the high-sensitivity measurements of methane and other trace gases.