Optics and Precision Engineering, Volume. 31, Issue 21, 3221(2023)

Hyperspectral images feature extraction and classification based on fractional differentiation

Jing LIU1,*... Yang LI1 and Yi LIU2 |Show fewer author(s)
Author Affiliations
  • 1School of Electronic Engineering, Xi’an University of Posts and Telecommunications, Xi'an702, China
  • 2School of Electronic Engineering, Xidian University, Xi'an710071, China
  • show less
    Figures & Tables(26)
    Eight-direction fractional differential mask
    Process of extracting hyperspectral image feature using fractional differentiation
    Grayscale images of Pavia university and its SpaFD feature
    Mean SpaFD images of four datasets
    Flow chart of constructing spectral-spatial joint criterion
    Relationship between proposed spectral-spatial joint criterion and order of SpaFD feature
    Classification maps of Indian Pines using 3DCNN under 5×5 fractional differential mask for feature extraction
    Classification maps of Botswana using 3DCNN under 5×5 fractional differential mask for feature extraction
    Classification maps of Pavia University using 3DCNN under 5×5 fractional differential mask for feature extraction
    Classification maps of Salinas using 3DCNN under 5×5 fractional differential mask for feature extraction
    [in Chinese]
    • Table 1. Convolutional kernel size settings for three networks

      View table
      View in Article

      Table 1. Convolutional kernel size settings for three networks

      网络模型I1Conv2Conv3Conv4Conv5FC6FC7FC8
      3DCNN5×5×N7×7×33×3×3----
      3DCNNPCA11×11×NP5×5×73×3×53×3×33×3×3
      HybridSN25×25×NP3×3×73×3×53×3×33×3
    • Table 2. Number of output feature maps for three network models

      View table
      View in Article

      Table 2. Number of output feature maps for three network models

      网络模型I1Conv2Conv3Conv4Conv5FC6FC7FC8
      3DCNN124--C--
      3DCNNPCA18163264256128C
      HybridSN18163264256128C
    • Table 3. Results of classical classifiers with 20% training samples

      View table
      View in Article

      Table 3. Results of classical classifiers with 20% training samples

      ClassifierIndian PinesBotswana
      SVMMDLRSVMMDLR
      Spe-Spa75.67±0.8949.69±1.2861.45±1.0873.52±1.1476.43±1.4687.31±1.19
      SpaFD5✕574.21±0.4647.87±1.6357.70±0.7360.80±2.2867.21±1.4979.81±1.74
      SpaFD-Spe-Spa5✕575.27±0.4648.12±0.8658.79±1.1768.45±1.4369.67±1.3584.80±1.41
      LBP-Spe-Spa70.40±0.2548.72±1.9461.27±1.3968.24±1.0276.00±0.9785.65±1.43
      ClassifierPavia UniversitySalinas
      SVMMDLRSVMMDLR
      Spe-Spa79.75±0.0759.67±0.4577.91±0.7582.68±0.7875.56±0.4585.01±1.24
      SpaFD5✕577.92±0.1358.56±0.4774.78±0.4178.98±0.1973.47±0.1883.70±0.85
      SpaFD-Spe-Spa5✕579.82±0.1558.70±0.4277.00±0.5581.51±0.6573.13±0.4184.06±1.13
      LBP-Spe-Spa75.83±0.1959.80±0.6077.07±0.6570.94±0.2374.50±0.3483.62±1.23
    • Table 4. Classification results of Indian Pines under 3% training samples

      View table
      View in Article

      Table 4. Classification results of Indian Pines under 3% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa73.91±5.0282.71±1.0486.65±1.10
      LBP60.58±4.3371.20±2.1175.52±1.02
      LBP-Spe-Spa71.75±2.6476.36±1.0580.46±0.88
      SpaFD3✕377.78±1.5684.10±0.6287.81±0.73
      SpaFD-Spe-Spa3✕377.81±2.2985.80±0.5988.94±0.67
      SpaFD5✕574.62±2.8783.48±0.8587.62±0.86
      SpaFD-Spe-Spa5✕575.44±2.6486.11±1.2288.89±0.83
      SpaFD7✕774.44±3.6482.97±0.6287.30±0.71
      SpaFD-Spe-Spa7✕774.83±2.9385.59±0.8088.52±0.73
    • Table 5. Classification results of Indian Pines under 5% training samples

      View table
      View in Article

      Table 5. Classification results of Indian Pines under 5% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa83.78±1.7389.61±0.5792.87±0.98
      LBP72.44±1.8978.55±1.0382.71±0.96
      LBP-Spe-Spa79.52±1.4683.41±0.8985.33±0.71
      SpaFD3✕384.47±1.1390.75±0.7793.44±0.86
      SpaFD-Spe-Spa3✕384.79±1.2292.04±0.4494.34±0.55
      SpaFD5✕584.43±1.9689.92±0.5293.46±0.72
      SpaFD-Spe-Spa5✕584.69±1.7590.42±0.4194.50±0.64
      SpaFD7✕783.87±1.0189.96±0.6493.39±0.78
      SpaFD-Spe-Spa7✕785.05±1.1492.32±0.3894.31±0.59
    • Table 6. Classification results of Indian Pines under 10% training samples

      View table
      View in Article

      Table 6. Classification results of Indian Pines under 10% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa88.93±0.8893.76±0.6897.99±0.45
      LBP79.03±0.9283.63±0.9787.76±0.50
      LBP-Spe-Spa84.89±1.0188.73±0.5489.98±0.39
      SpaFD3✕389.35±0.7793.87±0.2498.12±0.37
      SpaFD-Spe-Spa3✕389.49±0.8595.35±0.3598.86±0.40
      SpaFD5✕589.43±0.8594.17±0.4198.30±0.35
      SpaFD-Spe-Spa5✕589.50±1.1795.08±0.2199.01±0.42
      SpaFD7✕788.97±0.9293.99±0.2398.15±0.37
      SpaFD-Spe-Spa7✕789.39±0.7095.32±0.2898.89±0.41
    • Table 7. Classification results of Botswana dataset under 3% training samples

      View table
      View in Article

      Table 7. Classification results of Botswana dataset under 3% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa87.33±0.9590.69±0.8393.01±1.13
      LBP68.21±1.0180.32±1.3183.79±1.25
      LBP-Spe-Spa76.55±0.9684.65±0.9888.46±0.99
      SpaFD3✕387.85±0.9492.11±1.4994.08±0.98
      SpaFD-Spe-Spa3✕387.97±0.8796.29±0.9794.82±0.64
      SpaFD5✕587.89±1.1991.79±0.6994.11±1.05
      SpaFD-Spe-Spa5✕588.04±0.7196.31±0.6895.01±0.57
      SpaFD7✕788.01±1.5492.08±1.3193.85±1.03
      SpaFD-Spe-Spa7✕788.07±0.8996.01±0.8194.90±0.67
    • Table 8. Classification results of Botswana dataset under 5% training samples

      View table
      View in Article

      Table 8. Classification results of Botswana dataset under 5% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa90.77±1.6891.90±1.5094.05±1.16
      LBP71.36±1.4382.98±1.4685.01±0.89
      LBP-Spe-Spa83.49±0.9786.02±0.8889.32±0.92
      SpaFD3✕391.73±1.0192.64±1.4495.02±1.01
      SpaFD-Spe-Spa3✕392.10±0.9795.98±0.6297.90±0.76
      SpaFD5✕591.82±0.9092.34±1.7994.98±1.05
      SpaFD-Spe-Spa5✕592.08±0.4996.21±0.7397.77±0.80
      SpaFD7✕791.24±1.4191.92±0.7795.11±1.22
      SpaFD-Spe-Spa7✕792.18±0.9895.99±0.3998.05±0.94
    • Table 9. Classification results of Botswana dataset under 10% training samples

      View table
      View in Article

      Table 9. Classification results of Botswana dataset under 10% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa93.35±2.7697.77±0.4998.89±0.38
      LBP79.88±1.3685.46±0.5188.13±0.47
      LBP-Spe-Spa86.25±0.9790.04±0.3792.22±0.36
      SpaFD3✕394.97±0.7997.82±0.4599.00±0.31
      SpaFD-Spe-Spa3✕395.09±0.8999.35±0.2299.61±0.25
      SpaFD5✕594.53±0.6397.90±0.3898.98±0.29
      SpaFD-Spe-Spa5✕594.89±0.5999.30±0.2999.50±0.21
      SpaFD7✕794.46±0.7397.80±0.2799.02±0.25
      SpaFD-Spe-Spa7✕794.84±0.6099.49±0.1499.87±0.11
    • Table 10. Classification results of Pavia University under 3% training samples

      View table
      View in Article

      Table 10. Classification results of Pavia University under 3% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa88.51±3.4797.88±0.2998.42±0.52
      LBP76.73±3.5883.28±0.4087.56±0.58
      LBP-Spe-Spa80.61±0.8985.01±0.3390.31±0.55
      SpaFD3✕390.77±1.0998.00±0.2598.56±0.36
      SpaFD-Spe-Spa3✕391.46±0.6498.26±0.3698.88±0.26
      SpaFD5✕590.92±0.6197.96±0.3098.67±0.43
      SpaFD-Spe-Spa5✕591.86±0.7398.27±0.4598.93±0.25
      SpaFD7✕790.90±1.4998.04±0.5298.69±0.45
      SpaFD-Spe-Spa7✕791.75±0.5898.25±0.4399.01±0.37
    • Table 11. Classification results of Pavia University under 5% training samples

      View table
      View in Article

      Table 11. Classification results of Pavia University under 5% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa92.71±0.9198.47±0.2599.08±0.15
      LBP83.32±0.9587.12±0.2991.71±0.33
      LBP-Spe-Spa86.41±0.6389.90±0.3192.89±0.16
      SpaFD3✕393.46±0.4998.89±0.1299.14±0.20
      SpaFD-Spe-Spa3✕393.83±0.3598.97±0.3099.35±0.13
      SpaFD5✕593.06±0.8498.49±0.2499.17±0.21
      SpaFD-Spe-Spa5✕593.90±0.4298.79±0.2299.44±0.10
      SpaFD7✕792.88±1.2098.75±0.2899.18±0.20
      SpaFD-Spe-Spa7✕793.19±1.0199.06±0.1099.46±0.11
    • Table 12. Classification results of Pavia University under 10% training samples

      View table
      View in Article

      Table 12. Classification results of Pavia University under 10% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa94.82±0.5699.45±0.1199.70±0.08
      LBP86.51±0.5490.46±0.1593.01±0.11
      LBP-Spe-Spa89.44±0.4291.68±0.1094.97±0.08
      SpaFD3✕395.06±0.2599.54±0.1199.73±0.10
      SpaFD-Spe-Spa3✕395.32±0.3399.60±0.1299.79±0.07
      SpaFD5✕594.91±1.0199.49±0.0899.74±0.08
      SpaFD-Spe-Spa5✕595.19±0.4599.56±0.0499.81±0.11
      SpaFD7✕794.81±0.2899.50±0.1999.73±0.07
      SpaFD-Spe-Spa7✕794.97±0.5499.50±0.0599.80±0.06
    • Table 13. Classification results of Salinas dataset under 3% training samples

      View table
      View in Article

      Table 13. Classification results of Salinas dataset under 3% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa89.86±1.6397.17±0.4899.71±0.18
      LBP81.28±1.5588.32±0.5090.12±0.15
      LBP-Spe-Spa85.96±1.4689.85±0.4191.35±0.10
      SpaFD3✕391.83±1.6797.95±0.4599.73±0.20
      SpaFD-Spe-Spa3✕394.11±1.6498.48±0.4599.74±0.09
      SpaFD5✕590.25±1.5897.80±0.3299.75±0.14
      SpaFD-Spe-Spa5✕592.66±1.5498.32±0.3699.79±0.08
      SpaFD7✕790.75±1.9197.24±0.2599.75±0.18
      SpaFD-Spe-Spa7✕793.34±1.6197.99±0.2299.77±0.06
    • Table 14. Classification results of the Salinas dataset under 5% training samples

      View table
      View in Article

      Table 14. Classification results of the Salinas dataset under 5% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa90.81±1.2097.96±0.3099.80±0.09
      LBP82.86±1.4389.93±0.3591.84±0.10
      LBP-Spe-Spa86.75±0.8890.72±0.2493.44±0.07
      SpaFD3✕393.17±1.5698.45±0.3299.83±0.09
      SpaFD-Spe-Spa3✕395.99±0.4898.95±0.2699.88±0.07
      SpaFD5✕591.40±2.0398.60±0.3599.85±0.11
      SpaFD-Spe-Spa5✕595.34±0.9499.02±0.2299.89±0.07
      SpaFD7✕792.48±0.9798.58±0.3899.83±0.14
      SpaFD-Spe-Spa7✕795.15±0.8999.00±0.2899.87±0.10
    • Table 15. Classification results of Salinas dataset under 10% training samples

      View table
      View in Article

      Table 15. Classification results of Salinas dataset under 10% training samples

      Model

      3DCNN

      AOA(%)±SD(%)

      3DCNNPCA

      AOA(%)±SD(%)

      HybridSN

      AOA(%)±SD(%)

      Spe-Spa92.55±0.9799.60±0.1299.95±0.04
      LBP85.27±1.0294.43±0.2495.21±0.09
      LBP-Spe-Spa89.46±0.5095.01±0.1496.63±0.05
      SpaFD3✕395.42±1.0199.64±0.2599.95±0.09
      SpaFD-Spe-Spa3✕396.79±0.2899.71±0.1599.97±0.05
      SpaFD5✕593.76±1.1199.61±0.2799.96±0.03
      SpaFD-Spe-Spa5✕596.61±0.4699.65±0.1799.97±0.04
      SpaFD7✕793.82±2.4099.60±0.3699.94±0.10
      SpaFD-Spe-Spa7✕796.72±0.4099.62±0.2099.96±0.06
    Tools

    Get Citation

    Copy Citation Text

    Jing LIU, Yang LI, Yi LIU. Hyperspectral images feature extraction and classification based on fractional differentiation[J]. Optics and Precision Engineering, 2023, 31(21): 3221

    Download Citation

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

    Category:

    Received: May. 8, 2023

    Accepted: --

    Published Online: Jan. 5, 2024

    The Author Email: LIU Jing (zyhalj1975@163.com)

    DOI:10.37188/OPE.20233121.3221

    Topics