Spectroscopy and Spectral Analysis, Volume. 41, Issue 1, 94(2021)

Research of Terahertz Time-Domain Spectral Identification Based on Deep Learning

Qi-feng HU* and Jian CAI
Author Affiliations
  • Brainware Terahertz Information Technology Co., Ltd., Hefei 230088, China
  • show less
    Figures & Tables(11)
    Absorption spectra of five acids
    (a) Absorption spectra of Vitamin B2 of different concentrations; (b) Absorption spectra of Vitamin B2 in different humidity
    Result of S-G filtering
    RNN classification network structure
    CNN classification network structure
    Spectral images of 12 materials
    The training process of RNN network
    The training process of CNN network
    (a) Comparative experiment of concentrations of samples; (b) Comparative experiment of data dimensions
    • Table 1. Performance comparison of three algorithms

      View table
      View in Article

      Table 1. Performance comparison of three algorithms

      算法训练准确率/%测试准确率/%测试耗时/ms时间复杂度测试环境
      RNN97.697.5<1O(1)GPU: Nvidia GeForce RTX2080Ti
      CNN99.999.6<1O(1)GPU: Nvidia GeForce RTX2080Ti
      k-NN10087.6>100O(n)CPU
    • Table 2. The effect of data preprocessing on performance of RNN and CNN

      View table
      View in Article

      Table 2. The effect of data preprocessing on performance of RNN and CNN

      数据预处理RNN/%CNN/%
      78.783.4
      S-G滤波95.199.6
    Tools

    Get Citation

    Copy Citation Text

    Qi-feng HU, Jian CAI. Research of Terahertz Time-Domain Spectral Identification Based on Deep Learning[J]. Spectroscopy and Spectral Analysis, 2021, 41(1): 94

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Research Articles

    Received: Nov. 15, 2019

    Accepted: --

    Published Online: Apr. 8, 2021

    The Author Email: Qi-feng HU (fengmaomao1991@126.com)

    DOI:10.3964/j.issn.1000-0593(2021)01-0094-06

    Topics