基于深度学习的波长调制吸收光谱非线性层析重建研究

Deep Learning Method for Nonlinear Absorption Tomography with Calibration-Free Wavelength Modulation Spectroscopy

本文基于深度学习算法,开展了免标定波长调制吸收光谱非线性层析重建研究.采用频分复用技术,构建吸收光谱非线性层析重建方案,使用5条吸收谱线、2个投影角度共计20条投影光线.以二维温度场重建为例,在无噪声时温度场平均重建误差为2.85%,且在不同噪声水平和训练样本数目条件下,该方案可实现高精度温度和浓度场二维重建.本工作首次从算法理论层面实现了二维温度、浓度分布重建,将为实际燃烧环境中非线性层析重建实验提供指导.

This paper demonstrates a new method for solving nonlinear tomographic problems,combining calibration-free wavelength modulation spectroscopy(CF-WMS)with deep learning network.A number of laser diodes covering five H_(2)O absorption lines are combined using a wavelength multiplexer,each modulated at a different frequency.Two projections,each with 10 parallel beams,are assumed.The average reconstruction error of temperature distribution is 2.85%in the absence of noise.High-fidelity reconstruction can be achieved for both temperature and H_(2)O concentration distributions at various noise levels and numbers of dataset.This is the first time,to the best of the authors’knowledge,that deep learning network has been applied to WMS-based nonlinear tomography.The results provide guidelines for accurate reconstruction of temperature and concentration distributions in practical combustion environments.