• Spectroscopy and Spectral Analysis
  • Vol. 42, Issue 8, 2353 (2022)
Chun-ling WANG1,*, Kai-yuan SHI1,1; 2;, Xing MING3,3; *;, Mao-qin CONG3,3;..., Xin-yue LIU3,3; and Wen-ji GUO3,3;|Show fewer author(s)
Author Affiliations
  • 11. School of Information Science and Technology, Beijing Forestry University, Beijing 100083, China
  • 33. Nanjing Institute of Software Technology, Institute of Software Chinese Academy of Sciences, Nanjing 210049, China
  • show less
    DOI: 10.3964/j.issn.1000-0593(2022)08-2353-06 Cite this Article
    Chun-ling WANG, Kai-yuan SHI, Xing MING, Mao-qin CONG, Xin-yue LIU, Wen-ji GUO. A Comparative Study of the COD Hyperspectral Inversion Models in Water Based on the Maching Learning[J]. Spectroscopy and Spectral Analysis, 2022, 42(8): 2353 Copy Citation Text show less
    The original spectral reflectance curve of water samples
    Fig. 1. The original spectral reflectance curve of water samples
    Spectral profiles of water samples after preprocess(a): SG smoothing; (b): MSC; (c): SG smoothing and MSC
    Fig. 2. Spectral profiles of water samples after preprocess
    (a): SG smoothing; (b): MSC; (c): SG smoothing and MSC
    Relationship between the number of decision trees and model MSE on training sample(a): Random forest; (b): Adaboost; (c): XGBoost
    Fig. 3. Relationship between the number of decision trees and model MSE on training sample
    (a): Random forest; (b): Adaboost; (c): XGBoost
    Sccetterplots of XGBoost inversion model based on different preprocessing methods(a): Original data; (b): MSC; (c): SG smoothing; (d): SG smoothing and MSC
    Fig. 4. Sccetterplots of XGBoost inversion model based on different preprocessing methods
    (a): Original data; (b): MSC; (c): SG smoothing; (d): SG smoothing and MSC
    The variancecontribution rate of the first ten principal components about PCA
    Fig. 5. The variancecontribution rate of the first ten principal components about PCA
    数据集数量均值/
    (mg·L-1)
    最小值/
    (mg·L-1)
    最大值/
    (mg·L-1)
    SD/
    (mg·L-1)
    训练集1 238134.854.2216.725.0
    预测集310133.874.8258.624.3
    总样本集1 548134.654.2258.624.8
    Table 1. Results of chemical oxgen demand (COD) statictical value of samples
    机器学习
    模型
    测试集指标训练时间
    /s
    R2RMSE/(mg·L-1)RPD
    多元线性0.4818.51.31.7
    随机森林0.8510.02.4146
    AdaBoost0.877.63.2194
    XGBoost0.917.13.498.9
    Table 2. The results of machine learning model based on orginal data
    机器学习
    模型
    测试集指标训练时间
    /s
    R2RMSE/(mg·L-1)RPD
    多元线性0.328.30.862
    随机森林0.8410.02.4188
    AdaBoost0.888.32.9165
    XGBoost0.917.03.570
    Table 3. The results of machine learning model based on data processed by SG smoothing
    机器学习
    模型
    测试集指标训练时间
    /s
    R2RMSE/(mg·L-1)RPD
    多元线性0.2129.10.842.2
    随机森林0.8210.32.4124
    AdaBoost0.8310.32.4178
    XGBoost0.907.33.372
    Table 4. The results of machine learning model based on data processed by MSC
    机器学习
    模型
    测试集指标训练时间
    /s
    R2RMSE/(mg·L-1)RPD
    多元线性0.2129.10.842.2
    随机森林0.869.12.7145
    AdaBoost0.878.92.7163
    XGBoost0.927.13.472
    Table 5. The results of machine learning model based on data processed by SG smoothing and MSC
    特征提取方法测试集指标训练时间
    /s
    R2RMSE/(mg·L-1)RPD
    未经过PCA处理0.927.13.472
    PCA处理0.936.43.82.9
    Table 6. The result of the XGBoost model built based on the PCA method
    Chun-ling WANG, Kai-yuan SHI, Xing MING, Mao-qin CONG, Xin-yue LIU, Wen-ji GUO. A Comparative Study of the COD Hyperspectral Inversion Models in Water Based on the Maching Learning[J]. Spectroscopy and Spectral Analysis, 2022, 42(8): 2353
    Download Citation