用于TDLAS二氧化碳气体检测的神经网络滤波方法

doi:10.3964/j.issn.1000-0593(2025)06-1514-07

Abstract
References
Related Citation (15)

Download: PDF (6702 KB)
Export: BibTeX | EndNote (RIS)

Abstract Carbon dioxide (CO₂) gas detection has important research significance in various fields, such as environmental monitoring, agricultural production, and microbial detection. Tunable diode laser absorption spectroscopy(TDLAS/WMS) based on wavelength modulation detection systems has become an important means of precision gas detection due to its significant advantages of high sensitivity, low cost, non-invasiveness, and real-time monitoring. However, the system is susceptible to interference from various environmental noises, significantly impacting gas detection accuracy and stability. The commonly used traditional time-frequency analysis algorithm cannot effectively filter out the low-frequency signal noise coupled with the absorption signal, which will interfere with the subsequent gas concentration retrieval task. With their powerful feature mapping capabilities, deep learning algorithms can project signals into a new feature space, learn the distribution of spectral signal background structures, and thus overcome the limitations of time-frequency domain filtering algorithms. Therefore, a deep learning-based TDLAS carbon dioxide gas detection filtering algorithm (TGDF) is proposed to reduce the influence of full-frequency noise in the gas detection system and improve the accuracy of gas measurements. The TGDF takes a fully connected neural network as the infrastructure and adds sampling blocks to remove noise in the feature domain; in addition, the singular value decomposition is introduced to further adjust the harmonic signals. The model is trained, tested, and tuned by simulating different concentrations of CO₂ absorption spectra with noise under experimental conditions, and the model performance is tested on the experimental dataset. In the simulation experiments, the average signal-to-noise ratio of the TGDF-filtered spectra increased by a factor of 3.05 from 7.34 dB to 22.41 dB. It kept the lowest noise residuals in the frequency domain. In real experiments, there is a good linear relationship between the second harmonic amplitude and preset concentrations of CO₂ (R²=0.998); the average absolute error (MAE) of five CO₂ detections is divided into 0.27%, 0.20%, 0.23%, 0.28% and 0.32%. Compared with the commonly used filtering algorithms such as EMD, SG, Wavelet transform, and MLP neural networks in these two datasets, the TGDF showed the best performance in suppressing systematic noises of different frequencies and phases. The results fully proved that TGDF could effectively reduce the systematic noise of each frequency band in the harmonic signal of gas detection and improve the accuracy and stability of TDLAS CO₂ concentration detection, which provides a feasible technical means for high-sensitivity measurement of CO₂ and other trace gases.

Key words：Tunable diode laser absorption spectroscopy; Carbon dioxide gas detection; Filtering algorithms; Neural network filtering algorithms

Received: 2024-08-09 Accepted: 2024-12-10

ZTFLH:

O433

Corresponding Authors: JIANG Chen-yu E-mail: jiangcy@sibet.ac.cn

Cite this article:

ZHAO Yu-ying,WANG Le,HUANG Tian-he, et al. Neural Network Filtering Method for Carbon Dioxide Gas Detection in TDLAS[J]. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(06): 1514-1520.

URL:

https://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2025)06-1514-07 OR https://www.gpxygpfx.com/EN/Y2025/V45/I06/1514

[1] DAI Tong-xin, WANG Biao, LIAN Hou-quan, et al(戴童欣, 王彪, 连厚泉, 等). Laser Journal(激光杂志), 2021, 42(4): 41.
[2] JIN Jiang-tie, LI Yang, LI Rong-jun, et al(金奖铁, 李扬, 李荣俊, 等). Plant Physiology Journal(植物生理学报), 2019, 55(5): 558.
[3] Fodor N, Challinor A, Droutsas I, et al. Plant and Cell Physiology, 2017, 58(11): 1833.
[4] Shao Jie, Xiang Jindong, Axner O, et al. Applied Optics, 2016, 55(9): 2339.
[5] Sun Jing, Song Yuxiao, Shi Dongxin, et al. Processes, 2022, 10(8): 1450.
[6] Fu Bo, Zhang Chenghong, Lü Wenhao, et al. Applied Spectroscopy Reviews, 2022, 57(2): 112.
[7] LI Jin-yi, LI Lian-hui, ZHAO Shuo, et al(李金义，李连辉, 赵烁, 等). Laser & Optoelectronics Progress(激光与光电子学进展), 2022, 59(13): 65.
[8] LIU Zhao-hai, AN Xin-chen, TAO Zhi, et al(刘兆海, 安昕辰, 陶治, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2024, 44(2): 359.
[9] Lins B, Zinn P, Engelbrecht R, et al. Applied Physics B: Lasers and Optics, 2010, 100(2): 367.
[10] Li Jingsong, Yu Benli, Zhao Weixiong, et al. Applied Spectroscopy Reviews, 2014, 49(8): 666.
[11] TIAN Chuan, ZOU Li-chang, RUAN Bin, et al(田川，邹丽昌，阮斌，等). Chinese Journal of Quantum Electronics(量子电子学报), 2021, 38(5): 661.
[12] Zhang Guosheng, Hao He, Wang Yichen, et al. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2021, 263: 120187.
[13] FENG Shi-ling, CUI Qi, GUO Xin-qian, et al(冯仕凌，崔琪，郭心骞，等). Journal of Atmospheric and Environmental Optics(大气与环境光学学报), 2022, 17(3): 328.
[14] Li Chuanliang, Guo Xinqian, Ji Wenhai, et al. Optical and Quantum Electronics, 2018, 50: 275.
[15] Zhang Tianyu, Kang Jiawen, Meng Dezhuang, et al. Sensors, 2018, 18(12): 4295.
[16] Sun Jiachen, Chang Jun, Wei Yubin, et al. Analytical Chemistry, 2022, 94(41): 14290.
[17] Wang Senyuan, Yang Shicheng, Zhu Shouzheng, et al. Optics Express, 2024, 32(7): 11849.
[18] Luo Qiwu, Zhou Jian, Li Weichuang, et al. IEEE Sensors Journal, 2022, 22(3): 2419.
[19] Zhao Peng, Ding Dan, Li Kaitai, et al. Optics Communications, 2024, 567: 130327.