Pressure calibration-free gas sensing based on transfer learning assisted wavelength modulation spectroscopy

Pressure fluctuations significantly affect the accuracy and stability of traditional wavelength modulation spectroscopy (WMS) technology. To solve this problem, a novel pressure calibration-free gas sensor based on convolutional neural network (CNN) assisted calibration-free wavelength modulation spectroscopy (CF-WMS) was proposed and experimentally demonstrated. The transfer learning (TL) was introduced to fine-tune the CNN pretrained on the simulated datasets, effectively addressing the issue of scarce experimental data. The performance of the proposed method was investigated and compared with the WMS technique, a sensitivity enhancement factor of 4.17 is achieved under the condition of atmospheric pressure fluctuations within a range of +/- 10 mbar. Experimental results confirm the feasibility of utilizing deep learning to enhance WMS performance, particularly in conditions characterized by pressure fluctuations. This work indicates the potential of combining WMS with deep learning to promote the research in gas absorption spectroscopy.