Trace gas sensor based on photoacoustic spectroscopy and deep learning nested U-shaped network (U-Net++)

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Artificial Intelligence

Background:

• Photoacoustic spectroscopy (PAS) is a sensitive gas detection technique.
• Conventional PAS systems face challenges with noise and long-term stability.
• Optimizing acoustic resonators and employing advanced signal processing are key to improving PAS performance.

Purpose of the Study:

• To develop a novel trace gas sensor with high sensitivity and stability.
• To integrate a rollar-type acoustic resonator with a U-Net++ deep learning network.
• To enhance methane (CH4) detection capabilities for various applications.

Main Methods:

• Finite element simulations were used to optimize a rollar-type acoustic resonator, increasing signal amplification by 21%.
• A nested U-shaped deep learning network (U-Net++) was employed for dynamic spectral data processing and noise suppression.
• Allan variance analysis was performed to assess sensor stability and minimum detectable concentration (MDC).

Main Results:

• The optimized resonator achieved a resonance frequency of 1040 Hz and a quality factor of 80.
• The U-Net++ architecture reduced the standard deviation of methane measurements by three orders of magnitude (12.91 ppm to 0.0138 ppm).
• The MDC for methane improved by over 8100 times, from 1.41 ppm to 0.173 ppb, demonstrating enhanced stability and sensitivity.

Conclusions:

• The synergistic integration of an optimized resonator and deep learning significantly advances PAS technology.
• The developed sensor exhibits high precision, robustness, and long-term stability for trace gas detection.
• This technology holds promise for applications in environmental monitoring, industrial safety, and medical diagnostics.