Y-Net: a dual-branch deep learning network for nonlinear absorption tomography with wavelength modulation...

Area of Science:

• Spectroscopy
• Machine Learning
• Tomography

Background:

• Nonlinear tomographic problems present significant challenges in scientific measurement.
• Calibration-free wavelength modulation spectroscopy (CF-WMS) is a technique used for species concentration measurements.
• Deep learning offers potential for solving complex inverse problems in tomography.

Purpose of the Study:

• To introduce and evaluate a new method for solving nonlinear tomographic problems.
• To combine CF-WMS with a dual-branch deep learning network (Y-Net).
• To assess the performance of Y-Net in reconstructing temperature and H2O concentration fields.

Main Methods:

• Developed a dual-branch deep learning network (Y-Net).
• Investigated the principle of CF-WMS and the Y-Net architecture, training, and performance.
• Generated 20,000 random samples with Gaussian distributions for temperature and H2O concentration.
• Utilized the Non-uniformity Coefficient (NUC) to characterize field complexity.
• Simulated data with four projections, each having 24 parallel beams.

Main Results:

• Achieved average reconstruction errors of 1.55% for temperature and 2.47% for H2O concentration on a testing dataset of 2000 samples.
• Observed standard deviations of 0.46% for temperature and 0.75% for H2O concentration.
• Found that reconstruction errors increase linearly with increasing Non-uniformity Coefficient (NUC).
• Demonstrated superior noise immunity and computational efficiency compared to the simulated annealing algorithm.

Conclusions:

• The Y-Net, applied to WMS-based nonlinear tomography, is a novel approach for accurate field reconstruction.
• The method shows significant advantages over existing algorithms, particularly in complex environments.
• This technique enables real-time, in situ monitoring of practical combustion environments.