TDLAS signal denoising based on PCA-GASF-DCResNet

Area of Science:

• Spectroscopy
• Signal Processing
• Machine Learning

Background:

• External noise interferes with gas concentration measurements in tunable diode laser absorption spectroscopy (TDLAS).
• Second-harmonic signals in TDLAS are susceptible to noise, impacting measurement accuracy.
• Advanced signal processing techniques are needed to mitigate noise in TDLAS.

Purpose of the Study:

• To propose and evaluate a novel deep learning algorithm for denoising TDLAS second-harmonic signals.
• To improve the accuracy and reliability of gas concentration measurements by reducing noise interference.
• To enhance the stability and consistency of spectral features for better analysis.

Main Methods:

• Principal Component Analysis (PCA) for dimensionality reduction.
• Gramian Angular Summation Field (GASF) transformation to create 2D signal representations.
• A dual-channel residual network (ResNet) with multilevel feature fusion for signal denoising (PCA-GASF-DCResNet).

Main Results:

• The PCA-GASF-DCResNet algorithm demonstrated superior denoising performance, particularly under low signal-to-noise ratio (SNR) conditions.
• The method improved the consistency and stability of peak values and peak-to-valley ratios in spectral signals.
• Experimental results on CO2 signals showed effective noise suppression, leading to smoother second-harmonic signals and reduced baseline fluctuations.

Conclusions:

• The proposed PCA-GASF-DCResNet algorithm effectively mitigates external noise interference in TDLAS measurements.
• This approach facilitates accurate and reliable gas concentration inversion by enhancing signal quality.
• The developed method offers a robust solution for improving TDLAS performance in noisy environments.