Optics and Precision Engineering, Volume. 31, Issue 13, 1950(2023)

Cross-scene hyperspectral image classification combined spatial-spectral domain adaptation with XGBoost

Aili WANG1... Shanshan DING1, He LIU2, Haibin WU1,* and Yuji IWAHORI3 |Show fewer author(s)
Author Affiliations
  • 1Heilongjiang Province Key Laboratory of Laser Spectroscopy Technology and Application, College of measurement and control technology and communication Engineering, Harbin University of Science and Technology, Harbin 50080, China
  • 2State Grid Heilongjiang Electric Power Co., Ltd, Integrated data center, Harbin 150010, China
  • 3Department of Computer Science, Chubu University, Aichi 487-8501, Japan
  • show less
    Figures & Tables(17)
    SSDAX model architecture
    Spatial-spectral attention model
    Comparison of two different convolution procedure
    Schematic diagram of large convolution kernel decomposition
    False-color map and ground-truth map of Pavia dataset
    False-color map and ground-truth map of Indiana dataset
    Comparison of feature distribution before and after DA on Pavia dataset
    Comparison of feature distribution before and after DA on Indiana dataset
    Classification results of Pavia dataset
    Classification results of Indiana dataset
    • Table 1. Land class details in Pavia dataset

      View table
      View in Article

      Table 1. Land class details in Pavia dataset

      No.CategorySourceTarget
      C1Trees2662 424
      C2Asphalt2661 704
      C3Parking lot265287
      C4Bitumen206685
      C5Meadow2731 251
      C6Soil2131 475
    • Table 2. Land class details in Indiana dataset

      View table
      View in Article

      Table 2. Land class details in Indiana dataset

      No.CategorySourceTarget
      C1Concrete/ Asphalt4 8672 942
      C2Corn cleanTill9 8226 029
      C3Corn cleanTill EW11 4147 999
      C4Orchard5 1061 562
      C5Soybeans cleanTill4 7314 792
      C6Soybeans cleanTill EW2 9961 638
      C7Wheat3 22310 739
    • Table 3. Parameters of spatial-spectral attention model

      View table
      View in Article

      Table 3. Parameters of spatial-spectral attention model

      Layer nameOutput shapeFilter sizePaddingDilationGroups
      Input5×5×ch----
      DoConv5×5×200(1, 1)(0, 0)(1, 1)1
      BN5×5×200----
      ReLu5×5×200----
      Dropout5×5×200----
      Avgpool1×1×200(5, 5)(0, 0)(1, 1)-
      DW-Conv1×1×200(5, 5)(2, 2)(1, 1)200
      DW-D-Conv1×1×200(7, 7)(9, 9)(3,3)200
      1×1 Conv1×1×200(1, 1)(0, 0)(1, 1)1
      FC128----
    • Table 4. Comparison of classification accuracy of different methods on Pavia dataset

      View table
      View in Article

      Table 4. Comparison of classification accuracy of different methods on Pavia dataset

      MethodC1C2C3C4C5C6OA/%AA/%K×100Time/s
      RBF-SVM79.5880.1981.5585.9284.1274.9779.3279.4077.3466.84
      EMP-SVM80.4080.7584.1788.1283.2978.5680.5380.8978.15330.90
      DCNN86.4784.4086.6690.4583.0482.5684.2083.9481.17277.76
      ED-DMM-UDA89.2086.5594.8092.2096.0982.1287.9888.5084.8539.31
      DCDA92.1494.3610081.3595.7881.9990.6390.0888.0042.74
      SSDAX94.1790.5899.1392.1594.2284.8191.6292.5189.3936.93
    • Table 5. Comparison of classification accuracy of different methods on Indiana dataset

      View table
      View in Article

      Table 5. Comparison of classification accuracy of different methods on Indiana dataset

      MethodC1C2C3C4C5C6C7OA/%AA/%K×100Time/s
      RBF-SVM59.4329.3242.9169.3022.0455.3051.9443.6147.1835.76158.26
      EMP-SVM60.8234.9034.0288.7730.9558.1075.6551.9854.7443.07424.65
      DCNN65.2239.1441.9882.9636.1261.1577.8756.1057.7846.26294.50
      ED-DMM-UDA72.2241.4836.0091.8133.3267.1580.6056.8260.3748.10170.24
      DCDA53.1825.0141.1593.9042.2980.7282.1661.6061.7953.00153.72
      SSDAX77.4454.2044.6391.1954.7864.3886.9865.9867.6658.57151.05
    • Table 6. Ablation experimental results on Pavia dataset

      View table
      View in Article

      Table 6. Ablation experimental results on Pavia dataset

      MethodC1C2C3C4C5C6OA/%AA/%K×100Time/s
      MDDUWK89.3691.4097.2397.2396.1484.7990.1291.0387.5437.84
      SSDA89.8191.5596.9787.9496.4384.7490.3891.2487.8638.47
      SDAX93.0489.1298.0992.0094.5985.8891.1692.1288.8235.72
      SSDAX94.1790.5899.1392.1594.2284.8191.6292.5189.3936.93
    • Table 7. Ablation experimental results on Indiana dataset

      View table
      View in Article

      Table 7. Ablation experimental results on Indiana dataset

      MethodC1C2C3C4C5C6C7OA/%AA/%K×100Time/s
      MDDUWK72.8644.8748.9291.7648.2268.1184.0863.4465.5555.72166.91
      SSDA71.8350.0648.3892.1548.9769.5383.9464.2566.4156.57156.83
      SDAX75.6053.1244.6288.5753.4966.0085.6065.0266.7157.42151.61
      SSDAX77.4454.2044.6391.1954.7864.3886.9865.9867.6658.57151.05
    Tools

    Get Citation

    Copy Citation Text

    Aili WANG, Shanshan DING, He LIU, Haibin WU, Yuji IWAHORI. Cross-scene hyperspectral image classification combined spatial-spectral domain adaptation with XGBoost[J]. Optics and Precision Engineering, 2023, 31(13): 1950

    Download Citation

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

    Category: Information Sciences

    Received: Sep. 27, 2022

    Accepted: --

    Published Online: Jul. 26, 2023

    The Author Email: WU Haibin (woo@hrbust.edu.cn)

    DOI:10.37188/OPE.20233113.1950

    Topics