Laser & Optoelectronics Progress, Volume. 60, Issue 15, 1530002(2023)

Typical Feature Classification and Identification Method Based on Hyperspectral Data

Da Xu, Jun Pan, Lijun Jiang*, and Yu Cao
Author Affiliations
  • Key College of Geo-Exploration Science and Technology, Jilin University, Changchun 130026, Jilin, China
  • show less
    Figures & Tables(25)
    Surface states of four types of features. (a) Soybean; (b) corn; (c) rice; (d) bare soil
    Average reflectance spectral curves of four types of features in the range of 350-2500 nm
    Spectral curves of some samples of four types of features from 350-1800 nm (20 bars). (a) Soybean; (b) corn; (c) rice; (d) bare soil
    SPA feature band selection results. (a) RMSE; (b) average spectral reflectance
    Distribution of feature sample points. (a) 410 nm and 542 nm; (b) 410 nm and 714 nm; (c) 410 nm and 856 nm; (d) 410 nm and 1423 nm; (e) 410 nm and 1475 nm; (f) 410 nm and 1712 nm
    Structure diagrams. (a) 1DCNN; (b) 1DCNN-SPA
    1DCNN model training results. (a) Loss; (b) classification accuracy
    1DCNN-SPA model training results. (a) Loss; (b) classification accuracy
    LSTM architecture
    Structure diagrams. (a) LSTM; (b) LSTM-SPA
    LSTM model training results. (a) Loss; (b) classification accuracy
    LSTM-SPA model training results. (a) Loss; (b) classification accuracy
    Overall classification accuracy of different models with different wave sets
    Different model accuracy metrics
    Confusion matrix for different model classifications. (a) BP; (b) KNN; (c) 1DCNN; (d) LSTM
    BP Spectral curves. (a) Soybeans classified correctly, soybeans misclassified into corn samples; (b) corns classified correctly, soybeans misclassified into corn samples
    KNN Spectral curves. (a) Corns classified correctly, corns misclassified into soybean samples; (b) soybeans classified correctly, corns misclassified into soybean samples
    Comparison of soybean misclassification into corn samples at different stages of BP with correct soybean and corn classification samples.(a) Stage 1(16); (b) stage 2(11); (c) stage 3(6); (d) stage 4(5)
    Comparison of corn misclassification into soybean samples at different stages of KNN with correct corn and soybean classification samples(No change in the fourth stage).(a) Stage 1(8); (b) stage 2(6); (c) stage 3(3)
    Overall classification accuracy of soybean and corn for four methods under different stage feature band sets
    • Table 1. Data set statistics

      View table

      Table 1. Data set statistics

      Data setNumber of dataSoybeanCornRiceBare soil
      Training set55716415132210
      Test set1404039853
    • Table 2. Classification results of different models

      View table

      Table 2. Classification results of different models

      Model typeLossClassification accuracy /%Model time consumption /min
      1DCNN0.1968297.1467.85
      1DCNN-SPA0.2524895.7142.42
    • Table 3. Classification results of different models

      View table

      Table 3. Classification results of different models

      Model typeLossClassification accuracy /%Model time consumption /min
      LSTM0.0110099.2913.78
      LSTM-SPA0.1202098.5712.12
    • Table 4. Overall classification accuracy of different models with different sets of bands

      View table

      Table 4. Overall classification accuracy of different models with different sets of bands

      Model typeRemove very important band subsets /%Remove subset of most important bands /%Remove subset of less important bands /%Remove subset of more important bands /%8 feature band sets /%
      BP71.4372.8675.0080.7183.57
      KNN69.2970.7172.8677.8682.14
      1DCNN80.7182.5685.7189.2995.71
      LSTM82.1483.5787.1490.0098.57
    • Table 5. Classification accuracy of various types of features under 8 feature bands of different models

      View table

      Table 5. Classification accuracy of various types of features under 8 feature bands of different models

      Model TypeMapping accuracy /%User accuracy /%Overall accuracy /%Kappa coefficient
      SoybeanCornRiceBare soilSoybeanCornRiceBare soil
      BP55.0094.8762.5010095.6566.0710094.6483.570.7613
      KNN62.5076.9287.5010075.7668.1810094.6482.140.7415
      1DCNN87.5097.4410010097.2288.3710010095.710.9383
      LSTM97.5097.4410010097.5097.4410010098.570.9794
    Tools

    Get Citation

    Copy Citation Text

    Da Xu, Jun Pan, Lijun Jiang, Yu Cao. Typical Feature Classification and Identification Method Based on Hyperspectral Data[J]. Laser & Optoelectronics Progress, 2023, 60(15): 1530002

    Download Citation

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

    Category: Spectroscopy

    Received: Jun. 11, 2022

    Accepted: Jul. 22, 2022

    Published Online: Aug. 11, 2023

    The Author Email: Jiang Lijun (jlijun@jlu.edu.cn)

    DOI:10.3788/LOP222050

    Topics