Carbon Dioxide Sensing Based on Off-Axis Integrated Cavity Absorption Spectroscopy Combined with the Informer and...

Area of Science:

• Spectroscopy
• Sensor Technology
• Artificial Intelligence

Background:

• Off-axis integrated cavity output spectroscopy (OA-ICOS) offers high sensitivity for gas detection by increasing effective absorption path length.
• Traditional noise filtering methods in OA-ICOS systems exhibit low efficiency and limited feature extraction capabilities for complex spectral data.
• Deep learning models excel at extracting features from large-scale spectral data, enabling efficient and accurate analysis.

Purpose of the Study:

• To develop a CO2 sensor utilizing OA-ICOS combined with deep learning for improved spectral data processing and concentration prediction.
• To investigate the effectiveness of an 'informer' neural network for filtering CO2 spectral time series and enhancing signal-to-noise ratio (SNR).
• To evaluate the performance of a multilayer perceptron (MLP) model for direct spectral feature extraction and CO2 concentration prediction.

Main Methods:

• Implementation of an OA-ICOS system operating in the near-infrared (1.602 μm) spectral region for CO2 detection.
• Utilization of a radiofrequency (RF) noise source to mitigate cavity-mode noise and boost SNR in the OA-ICOS system.
• Application of an 'informer' neural network for time-series spectral data filtering, followed by MLP for feature extraction and concentration prediction.

Main Results:

• The 'informer' filtering method approximately doubled the SNR compared to traditional methods like Savitzky-Golay, Kalman, and wavelet threshold filtering.
• The MLP model achieved a significant improvement in the linear correlation coefficient (R^2) from 79.74% to 98.52% for CO2 concentration measurements.
• The CO2 sensor's detection limit was improved 3.79-fold, reaching 1.38 ppm at 224.4 s using the MLP model, compared to the absorption-peak-fitting method.

Conclusions:

• The integration of deep learning methods, specifically the 'informer' and MLP models, offers a powerful approach for spectral data processing in spectroscopy-based sensing.
• The proposed sensor system demonstrates superior performance in terms of SNR, accuracy, and detection limits for CO2 monitoring.
• This study highlights the potential of deep learning to advance the field of optical gas sensing and spectral analysis.