2.765

2022影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

仿生嗅觉感知系统气体识别和浓度估计模型

相洪涛 张文文 肖文鑫 王磊 王远西

相洪涛, 张文文, 肖文鑫, 王磊, 王远西. 仿生嗅觉感知系统气体识别和浓度估计模型. 自动化学报, xxxx, xx(x): x−xx doi: 10.16383/j.aas.c220689
引用本文: 相洪涛, 张文文, 肖文鑫, 王磊, 王远西. 仿生嗅觉感知系统气体识别和浓度估计模型. 自动化学报, xxxx, xx(x): x−xx doi: 10.16383/j.aas.c220689
Xiang Hong-Tao, Zhang Wen-Wen, Xiao Wen-Xin, Wang Lei, Wang Yuan-Xi. Gas recognition and concentration estimation model for bionic olfactory perception system. Acta Automatica Sinica, xxxx, xx(x): x−xx doi: 10.16383/j.aas.c220689
Citation: Xiang Hong-Tao, Zhang Wen-Wen, Xiao Wen-Xin, Wang Lei, Wang Yuan-Xi. Gas recognition and concentration estimation model for bionic olfactory perception system. Acta Automatica Sinica, xxxx, xx(x): x−xx doi: 10.16383/j.aas.c220689

仿生嗅觉感知系统气体识别和浓度估计模型

doi: 10.16383/j.aas.c220689
基金项目: 国家自然科学基金(62203307)资助
详细信息
    作者简介:

    相洪涛:山西大学自动化与软件学院硕士研究生. 主要研究方向为信号处理和深度学习. E-mail: xianghongtao0921@163.com

    张文文:南洋理工大学电气与电子工程学院研究员. 2021年获得同济大学博士学位. 主要研究方向为智能信息处理与模式识别算法, 仿生传感器检测技术与测量系统, 信号处理和深度学习. 本文通信作者. E-mail: wenwen.zhang@ntu.edu.sg

    肖文鑫:北京大学计算机学院博士研究生. 主要研究方向为软件工程和机器学习. E-mail: wenxin.xiao@stu.pku.edu.cn

    王磊:同济大学电子与信息工程学院教授. 主要研究方向为传感器检测技术与测量系统. E-mail: leiwang@tongji.edu.cn

    王远西:同济大学电子与信息工程学院博士研究生. 主要研究方向为传感器检测技术与测量系统. E-mail: 2010146@tongji.edu.cn

Gas Recognition and Concentration Estimation Model for Bionic Olfactory Perception System

Funds: Supported by National Natural Science Foundation of China (62203307)
More Information
    Author Bio:

    XIANG Hong-Tao Master student at the School of Automation and Software Engineering, Shanxi University. His research interest covers signal processing and deep learning

    ZHANG Wen-Wen Research fellow at the School of Electrical and Electronic Engineering, Nanyang Technological University. He received his Ph. D. degree from Tongji University in 2021. His research interest covers intelligent information processing and pattern recognition algorithms, bionic sensor detection technology and measurement systems, signal processing and deep learning. Corresponding author of this paper

    XIAO Wen-Xin Ph. D. candidate at the School of Computer Science, Peking University. His research interest covers software engineering and machine learning

    WANG Lei Professor at the College of Electronic and Information Engineering, Tongji University. His research interest covers sensor detection technology and measurement system

    WANG Yuan-Xi Ph. D. candidate at the College of Electronic and Information Engineering, Tongji University. His research interest covers sensor detection technology and measurement system

  • 摘要: 常用的气体检测模型需要使用气体传感器阵列响应信号的稳态值对气体进行种类识别和浓度估计, 而实际环境中, 气体一般处于动态变化的状态, 气体传感器阵列响应信号难以达到稳态值或长时间维持稳态状态. 针对上述问题, 提出了一种由动态小波残差卷积神经网络(Dynamic wavelet residual convolutional neural networks, DWRCNN)子模型和权重信号自注意力(Weighted signal self-attention, WSSA)子模型组成的气体检测模型. 该模型可以直接使用气体传感器阵列的原始动态响应信号对动态变化的气体进行成分识别, 并进一步对每种成分气体的浓度在线估计. 经自搭建的仿生嗅觉感知系统对模型的性能进行评估, 结果表明, 与常用气体识别模型相比, DWRCNN能获得接近于100%气体识别准确率, 且在线训练时间短, 收敛速度快; 与常用的气体浓度估计模型相比, WSSA 浓度估计模型能够大幅提高气体浓度估计精度, 并且能够同时对不同的气体都保持较高的气体浓度估计精度, 解决了动态环境中仿生嗅觉感知系统需要针对不同的气体选择不同的最优气体浓度估计模型的问题.
  • 图  1  DWRCNN-WSSA模型气体检测流程

    Fig.  1  Flow of gas detection by DWRCNN-WSSA model

    图  2  实验装置和平台((a)仿生嗅觉感知系统;(b)标准实验平台)

    Fig.  2  Experimental setup and platform ((a) Bionic olfactory sensing system; (b) Standard experimental platform)

    图  3  140 ppm CO传感器阵列动态响应信号曲线

    Fig.  3  Dynamic response signal curve of the sensor array for 140 ppm CO

    图  4  传感器阵列动态响应信号小波分解过程

    Fig.  4  Wavelet decomposition process of dynamic response signal of sensor array

    图  5  TGS2610在140 ppm CO下的动态响应信号曲线和相应的5层低频小波系数曲线

    Fig.  5  Dynamic response signal curve and corresponding 5-layer low-frequency wavelet coefficient curve at 140 ppm CO for TGS2610

    图  6  传感器阵列动态响应信号转换为小波系数图像过程

    Fig.  6  Process of converting the dynamic response signal of the sensor array into a wavelet coefficient map

    图  7  DWRCNN气体识别模型的结构

    Fig.  7  Structure of the DWRCNN gas identification mode

    图  8  计算复杂度图解

    Fig.  8  Illustration of computational complexity

    图  9  WSSA气体浓度估计模型的结构

    Fig.  9  Structure of the WSSA gas concentration estimation model

    图  10  不同模型的气体识别结果混淆矩阵

    Fig.  10  Confusion matrix of gas recognition results with different models

    图  11  不同模型气体识别的准确率曲线和损失函数曲线

    Fig.  11  Accuracy curve and loss function curve of gas recognition with different models

    图  12  PCA降维可视化

    Fig.  12  Visualization of PCA dimensionality reduction.

    图  13  不同模型的气体浓度估计误差箱式图

    Fig.  13  Error box plots of gas concentration estimation with different models

    图  14  SA模型和WSSA模型的气体浓度估计散点图

    Fig.  14  Scatter plots of gas concentration estimation for SA model and WSSA model

    图  15  气腔进气口位置和传感器阵列高度示意图

    Fig.  15  Diagram of gas cavity inlet position and sensor array height

    表  1  气体传感器阵列详细信息

    Table  1  Gas sensor array details

    ChannelSensor typeInc. Main sensitive gas types
    Channel 0MQ135WinsenNH3, H2S, C6H6
    Channel 1TGS813FIGAROCH4, 3CH3CH2CH3
    Channel 2TGS2611FIGAROCH4
    Channel 3TGS2610FIGAROCH3CH2CH3, C4H10
    Channel 4TGS2620FIGAROC2H6O, Organic solvents
    Channel 5TGS2600FIGARO H2, C2H6O
    Channel 6TGS2602FIGAROVOC, NH3, H2S, CH2O
    Channel 7MP503WinsenC2H6O, C4H10, CH2O
    下载: 导出CSV

    表  2  不同模型的气体识别准确率

    Table  2  Gas recognition accuracy of different models

    方法KNNSVMRFNB
    准确率95.74%96.45%95.74%92.91%
    方法BPNNCNNCapsNetDWRCNN
    准确率97.87%99.29%100%100%
    下载: 导出CSV

    表  3  CO浓度估计指标

    Table  3  Metrics of CO concentration estimation

    方法MAERMSEEVR$^2$
    BR6.5528.0940.9430.942
    SVM4.2587.0150.9630.957
    DT5.4728.0390.9490.943
    KNN5.0337.0750.9580.956
    RF4.7137.0740.9590.956
    Adaboost5.4777.6430.9500.949
    GBDT4.8177.0190.9600.957
    Bagging4.7607.0610.9590.956
    XGBoost4.6727.0350.9610.960
    OSA3.6304.2290.9870.986
    LSTM2.9343.8450.9880.988
    WS-LSTM2.3503.0350.9930.993
    SA2.9163.7560.9890.989
    WSSA2.0902.6460.9950.994
    下载: 导出CSV

    表  4  H$_2$浓度估计指标

    Table  4  Metrics of H$_2$ concentration estimation

    方法MAERMSEEVR$^2$
    BR16.09718.2840.6830.638
    SVM5.0346.9760.9550.947
    DT5.2068.9870.9210.913
    KNN6.3128.8650.9310.915
    RF5.0737.1570.9510.945
    Adaboost5.4417.2090.9520.944
    GBDT5.6878.4440.9310.923
    Bagging5.3467.6670.9400.936
    XGBoost5.5128.7240.9370.935
    OSA4.1555.3430.9770.977
    LSTM4.2645.3050.9740.973
    WS-LSTM3.7814.4570.9850.984
    SA4.1565.5600.9750.975
    WSSA2.3603.0280.9930.992
    下载: 导出CSV

    表  5  混合气体中CO浓度估计指标

    Table  5  Metrics of CO concentration estimation in the gas mixture

    方法MAERMSEEVR2
    BR20.13424.4870.5300.526
    SVM20.00924.6570.5190.519
    DT20.53725.7460.5150.476
    KNN21.23626.7010.4430.436
    RF18.52923.6140.5810.559
    Adaboost19.95025.0190.5260.505
    GBDT20.83026.1930.4810.457
    Bagging19.60825.2880.5310.494
    XGBoost15.93120.1010.6190.592
    OSA10.90913.5890.8600.859
    LSTM10.43914.0500.8560.849
    WS-LSTM7.95811.1880.9110.904
    SA9.20912.9580.8720.872
    WSSA6.0147.6160.9560.956
    下载: 导出CSV

    表  6  混合气体中H2浓度估计指标

    Table  6  Metrics of H2 concentration estimation in the gas mixture

    方法MAERMSEEVR2
    BR9.95612.3780.8970.897
    SVM8.00810.1060.9310.931
    DT11.32615.5030.8420.838
    KNN7.2979.6410.9370.937
    RF7.85210.8230.9220.921
    Adaboost9.12011.5820.9150.909
    GBDT7.76310.6220.9240.924
    Bagging8.01911.3940.9150.912
    XGBoost7.84010.0890.9320.931
    OSA8.88611.7200.9120.910
    LSTM7.7838.8480.9490.949
    WS-LSTM5.0956.8780.9690.969
    SA5.9067.7760.9600.960
    WSSA4.3186.3620.9740.973
    下载: 导出CSV

    表  7  气体识别准确率

    Table  7  Gas recognition accuracy

    气腔进气口位置ABC
    准确率100%100%100%
    传感器阵列摆放高度EFG
    准确率100%100%100%
    下载: 导出CSV

    表  8  气腔进气口位置不同时单一气体浓度估计的指标

    Table  8  Metrics for concentration estimation of single gas with different gas cavity inlet positions

    进气口位置气体种类MAERMSEEVR2
    ACO2.0902.6460.9950.994
    H22.3603.0280.9930.992
    BCO2.3263.0170.9940.994
    H22.2872.8980.9940.994
    CCO2.1852.8120.9950.994
    H22.4193.1770.9920.992
    下载: 导出CSV

    表  9  气腔进气口位置不同时混合气体浓度估计的指标

    Table  9  Metrics for concentration estimation of mixed gases with different gas cavity inlet positions

    进气口位置气体种类MAERMSEEVR2
    ACO6.0147.6160.9560.956
    H24.3186.3620.9740.973
    BCO5.6796.8990.9630.962
    H24.5626.7130.9730.973
    CCO5.8787.2560.9610.961
    H24.7856.8960.9720.972
    下载: 导出CSV

    表  10  传感器阵列摆放高度不同时单一气体浓度估计指标

    Table  10  Metrics for concentration estimation of single gas with different sensor array placement heights

    高度气体种类MAERMSEEVR2
    ECO2.0902.6460.9950.994
    H22.3603.0280.9930.992
    FCO2.2832.8780.9940.994
    H22.2262.8730.9940.994
    GCO2.3753.1220.9930.993
    H22.4513.1630.9920.992
    下载: 导出CSV

    表  11  传感器阵列摆放高度不同时混合气体浓度估计指标

    Table  11  Metrics for concentration estimation of mixed gases with different sensor array placement heights

    高度气体种类MAERMSEEVR2
    ECO6.0147.6160.9560.956
    H24.3186.3620.9740.973
    FCO6.3238.0120.9550.955
    H24.6196.9920.9720.972
    GCO6.2257.8960.9560.955
    H24.6737.1050.9720.972
    下载: 导出CSV

    表  12  模型的气体识别准确率

    Table  12  Gas recognition accuracy of model

    信号采集第1次第2次第3次
    准确率100%100%99.29%
    下载: 导出CSV

    表  13  单一气体浓度估计指标

    Table  13  Metrics for concentration estimation of single gas

    信号采集气体种类MAERMSEEVR2
    第1次CO2.0902.6460.9950.994
    H22.3603.0280.9930.992
    第2次CO2.5123.2830.9920.992
    H22.8143.6840.9910.991
    第3次CO2.9853.8720.9890.989
    H23.3504.1150.9880.987
    下载: 导出CSV

    表  14  混合气体浓度估计指标

    Table  14  Metrics for concentration estimation of mixed gases

    信号采集气体种类MAERMSEEVR2
    第1次CO6.0147.6160.9560.956
    H24.3186.3620.9740.973
    第2次CO6.7118.8130.9410.940
    H25.1576.9720.9670.967
    第3次CO7.0169.4370.9340.934
    H25.8157.6540.9620.962
    下载: 导出CSV
  • [1] Jing Y Q, Meng Q H, Qi P F, Cao M L, Zeng M, Ma S G. A bioinspired neural network for data processing in an electronic nose. IEEE Transactions on Instrumentation and Measurement, 2016, 65(10): 2369-2380. doi: 10.1109/TIM.2016.2578618
    [2] Persaud K, Dodd G. Analysis of discrimination mechanisms in the mammalian olfactory system using a model nose. Nature, 1982, 299(5881): 352-355. doi: 10.1038/299352a0
    [3] Jiang X, Jia P F, Luo R D, Deng B, Duan S K, Jia Yan. A novel electronic nose learning technique based on active learning: EQBC-RBFNN. Sensors and Actuators B: Chemical, 2017, 249: 533-541. doi: 10.1016/j.snb.2017.04.072
    [4] Zhang L, Zhang D. Efficient solutions for discreteness, drift, and disturbance (3D) in electronic olfaction. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2018, 48(2): 242-254. doi: 10.1109/TSMC.2016.2597800
    [5] Zhang L, Tian F C, Kadri C, Guangshu Pei, Li H J, Pan L. Gases concentration estimation using heuristics and bio-inspired optimization models for experimental chemical electronic nose. Sensors and Actuators B: Chemical, 2011, 160(1): 760-770. doi: 10.1016/j.snb.2011.08.060
    [6] Buma A I, Muller M, de Vries R, Sterk P J, van der Noort V, Wolf-Lansdorf M, et al. eNose analysis for early immunotherapy response monitoring in non-small cell lung cancer. Lung Cancer, 2021, 160: 36-43. doi: 10.1016/j.lungcan.2021.07.017
    [7] Lee J M, Choi E J, Chung J H, Lee K W, Lee Y, Kim Y J, et al. A DNA-derived phage nose using machine learning and artificial neural processing for diagnosing lung cancer. Biosensors and Bioelectronics, 2021, 194: 113567 doi: 10.1016/j.bios.2021.113567
    [8] Xu M, Wang J, Zhu L. Tea quality evaluation by applying E-nose combined with chemometrics methods. Journal of Food Science and Technology, 2021, 58: 1549-1561. doi: 10.1007/s13197-020-04667-0
    [9] Sanaeifar A, Li X L, He Y, Huang Z X, Zhan Z H. A data fusion approach on confocal Raman microspectroscopy and electronic nose for quantitative evaluation of pesticide residue in tea. Biosystems Engineering, 2021, 210: 206-222. doi: 10.1016/j.biosystemseng.2021.08.016
    [10] Ari D, Alagoz B B. An effective integrated genetic programming and neural network model for electronic nose calibration of air pollution monitoring application. Neural Computing and Applications, 2022, 34(15): 12633-12652. doi: 10.1007/s00521-022-07129-0
    [11] 高月, 宿翀, 李宏光. 一类基于非线性PCA和深度置信网络的混合分类器及其在PM2.5浓度预测和影响因素诊断中的应用. 自动化学报, 2018, 44(2): 318-329 doi: 10.16383/j.aas.2018.c160045

    GAO Yue, SU Chong, LI Hong-Guang. A kind of deep belief networks based on nonlinear features extraction with application to PM2.5 concentration prediction and diagnosis. Acta Automatica Sinica, 2018, 44(2): 318-329. doi: 10.16383/j.aas.2018.c160045
    [12] Patil S J, Duragkar N, Rao V R. An ultra-sensitive piezoresistive polymer nano-composite microcantilever sensor electronic nose platform for explosive vapor detection. Sensors and Actuators B: Chemical, 2014, 192: 444-451. doi: 10.1016/j.snb.2013.10.111
    [13] Yan K, Zhang D. Improving the transfer ability of prediction models for electronic noses. Sensors and Actuators B: Chemical, 2015, 220: 115-124. doi: 10.1016/j.snb.2015.05.060
    [14] Liao F, Yin S, Toney M F, Subramanian V. Physical discrimination of amine vapor mixtures using polythiophene gas sensor arrays. Sensors and Actuators B: Chemical, 2010, 150: 254-263. doi: 10.1016/j.snb.2010.07.006
    [15] Hu H, Yang X X, Guo X D, Khaliji K, Biswas S R, Javier García de Abajo F, et al. Gas identification with graphene plasmons. Nature Communication, 2019, 10(1): 1-7. doi: 10.1038/s41467-018-07882-8
    [16] Qian J H, Tian F C, Luo Y, Lu M C, Zhang A L. A novel multisensor detection system design for low concentrations of volatile organic compounds. IEEE Transactions on Industrial Electronics, 2021, 69(5): 5314-5324.
    [17] Hu J Z, Qu H M, Chang Y, Pang W, Zhang Q K, Liu J, et al. Miniaturized polymer coated film bulk acoustic wave resonator sensor array for quantitative gas chromatographic analysis. Sensors and Actuators B: Chemical, 2018, 274: 419-426. doi: 10.1016/j.snb.2018.07.162
    [18] 孟凡利, 季瀚洋, 苑振宇, 张华, 王稼鹏. 二氧化锡传感器对挥发性有机物的动态测试方法研究. 自动化学报, 2022, 48(3): 926-934

    MENG Fan-Li, JI Han-Yang, YUAN Zhen-Yu, ZHANG Hua, WANG Jia-Peng. Study on dynamic testing method of volatile organic compounds by tin dioxide sensor. Acta Automatica Sinica, 2022, 48(3): 926-934.
    [19] Xiong Y Z, Chen Y T, Chen C M, Wei X W, Xue Y Y, Wan H, et al. An odor recognition algorithm of electronic noses based on convolutional spiking neural network for spoiled food identification. Journal of the Electrochemical Society, 2021, 168(7): 077519. doi: 10.1149/1945-7111/ac1699
    [20] Wang S H, Chou T I, Tang K T. Using a hybrid deep neural network for gas classification. IEEE Sensors Journal, 2021, 21(5): 6401-6407. doi: 10.1109/JSEN.2020.3038304
    [21] Zhang L, Deng P L. Abnormal odor detection in electronic nose via self-expression inspired extreme learning machine. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2017, 49(10): 1922-1932.
    [22] He A X, Wei G F, Yu J, Tang Z N, Lin Z H, Wang P J. A novel dictionary learning method for gas identification with a gas sensor array. IEEE Transactions on Industrial Electronics, 2017, 64(12): 9709-9715. doi: 10.1109/TIE.2017.2748034
    [23] Sun H, Tian F C, Liang Z F, Sun T, Yu B, Yang S X, et al. Sensor array optimization of electronic nose for detection of bacteria in wound infection. IEEE Transactions on Industrial Electronics, 2017, 64(9): 7350-7358. doi: 10.1109/TIE.2017.2694353
    [24] Liu Y J, Meng Q H, Zhang X N. Data processing for multiple electronic noses using sensor response visualization. IEEE Sensors Journal, 2018, 18(22): 9360-9369. doi: 10.1109/JSEN.2018.2871599
    [25] Zhang L, Zhang D. Domain adaptation extreme learning machines for drift compensation in E-nose systems. IEEE Transactions on Instrumentation and Measurement, 2014, 64(7): 1790-1801.
    [26] Liu Y J, Meng Q H, Qi P F, Sun B, Zhu X S. Using spike-based bio-inspired olfactory model for data processing in electronic noses. IEEE Sensors Journal, 2017, 18(2): 692-702.
    [27] He K M, Zhang X Y, Ren S Q, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR). Las Vegas, USA: IEEE, 2016. 770−778.
    [28] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, et al. Attention is all you need. In: Proceedings of the 35th International Conference on Neural Information Processing Systems. State of California, USA: NIPS, 2017.5998-6008.
    [29] Qin Z, Sun W X, Deng H, Li D X, Wei Y S, Lv B H, et al. cosformer: Rethinking softmax in attention. In: Proceedings of the 10th International Conference on Learning Representations (ICLR). OpenReview.net, 2022.
    [30] Crowther P S, Cox R J. A method for optimal division of data sets for use in neural networks. In: Proceedings of the 9th International conference on knowledge-based and intelligent information and engineering systems. Melbourne, Australia: Springer, 2005. 1−7.
    [31] Zhang W W, Wang L, Chen J, Xiao W X, Bi X. A novel gas recognition and concentration detection algorithm for artificial olfaction. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1-14.
    [32] Zhang W W, Xiang H T, Wang Y X, Bi X, Zhang Y Z, Zhang P J, et al. A signal response visualization gas recognition algorithm based on a wavelet transform coefficient map-capsule network for artificial olfaction. IEEE Sensors Journal, 2022, 22(15): 14717-14726. doi: 10.1109/JSEN.2022.3184963
  • 加载中
计量
  • 文章访问数:  85
  • HTML全文浏览量:  47
  • 被引次数: 0
出版历程
  • 网络出版日期:  2023-10-30

目录

    /

    返回文章
    返回