Deep Learning-Enhanced Dual-Component Gas Sensor Based on Wavelength Modulation Spectroscopy.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Spectral overlap in gas mixtures complicates accurate detection.
• Wavelength Modulation Spectroscopy (WMS) is a sensitive technique for gas sensing.
• Simultaneous detection of multiple gases, like CO2 and CH4, is challenging.

Purpose of the Study:

• To develop a deep learning-enhanced dual-component gas sensor for simultaneous CO2 and CH4 detection.
• To address spectral overlap interference using advanced AI models.
• To enable accurate quantification of exhaled CO2 and CH4 for potential breath diagnostics.

Main Methods:

• Implemented a single-laser WMS system with 2f/1f signal detection.
• Utilized a Convolutional Neural Network (CNN) for concentration prediction (CPM).
• Employed Generative Adversarial Networks (GANs) for data augmentation to overcome limited experimental data.

Main Results:

• Achieved accurate simultaneous detection of CO2 and CH4 concentrations.
• Demonstrated high reliability with linear fitting against standard concentrations.
• Reported minimum detection limits of 17.34 ppm for CO2 and 3.52 ppb for CH4.
• Successfully measured exhaled CO2 and CH4 concentrations in practical application.

Conclusions:

• The deep learning-enhanced WMS sensor effectively overcomes spectral overlap for dual-component gas detection.
• The proposed method provides a feasible and reliable approach for simultaneous multicomponent gas measurement.
• This technology shows significant potential for non-invasive breath diagnosis.