Laser & Optoelectronics Progress, Volume. 59, Issue 2, 0210014(2022)

Hyperspectral Image Classification Based on Modified DenseNet and Spatial Spectrum Attention Mechanism

Xin Wang* and Yanguo Fan
Author Affiliations
  • College of Oceanography and Spatial Information, China University of Petroleum (East China), Qingdao , Shandong 266500, China
  • show less
    Figures & Tables(18)
    Modified three-dimensional convolution module
    Structure of the modified Dense_Layer
    Model of the spatial spectrum attention mechanism. (a) Channel attention mechanism; (b) spatial attention mechanism
    Structure of the Dense_Layer
    Structure of the MDSSAN model
    Indian Pines data set and label. (a) Indian Pines data set; (b) label
    Pavia University data set and label. (a) Pavia University data set; (b) label
    KSC data set and label. (a) KSC data set; (b) label
    Classification results of the Indian Pines data set. (a) Indian Pines data set; (b) label; (c) 2D_CNN; (d) 3D_CNN; (e) M3RCNN; (f) 3D_DenseNet; (g) MDSSAN
    Classification results of the Pavia University data set. (a) Pavia University data set; (b) label; (c) 2D_CNN;(d) 3D_CNN; (e) M3RCNN; (f) 3D_DenseNet; (g) MDSSAN
    Classification result diagram of the KSC data set. (a) KSC data set; (b) label; (c) 2D_CNN; (d) 3D_CNN; (e) M3RCNN; (f) 3D_DenseNet; (g) MDSSAN
    • Table 1. Number of samples of training set, verification set and test set selected from Indian Pines data set

      View table

      Table 1. Number of samples of training set, verification set and test set selected from Indian Pines data set

      No.CategoryTraining setVerification setTest set
      Total216411827311
      1alfalfa11434
      2corn-notill2841411009
      3corn-min167100572
      4corn5828163
      5grass/pasture10375320
      6grass/trees14870530
      7grass/pasture-mowed131125
      8hay-windrowed10651345
      9oats121223
      10soybeans-notill202111689
      11soybeans-min5092421737
      12soybeans-clean12673430
      13wheat5526163
      14woods250155902
      15building-grass-trees-drives8555291
      16stone-steel towers352878
    • Table 2. Number of samples of training set, verification set and test set selected from Pavia University data set

      View table

      Table 2. Number of samples of training set, verification set and test set selected from Pavia University data set

      No.CategoryTraining setVerification setTest set
      Total8514440829989
      1asphalt13066914637
      2meadows3730190213023
      3gravel4002251483
      4trees6263362114
      5sheets244155961
      6bare soil10334743540
      7bitumen282128941
      8bricks7144052587
      9shadows17992703
    • Table 3. Number of samples of training set, verification set and test set selected from KSC data set

      View table

      Table 3. Number of samples of training set, verification set and test set selected from KSC data set

      No.CategoryTraining setVerification setTest set
      Total11226233739
      1scrub17270522
      2willow swamp5032167
      3camping hammock5228185
      4slash pine6626172
      5oak/broadleaf3919118
      6hardwood4932166
      7swap232479
      8graminoid marsh8162312
      9spartina marsh10655386
      10cattail marsh8844302
      11salt marsh9058304
      12mud flats10366370
      13water203107656
    • Table 4. Classification results of the Indian Pines data set

      View table

      Table 4. Classification results of the Indian Pines data set

      Classification2D_CNN3D_CNNM3RCNN3D_DenseNetMDSSAN
      Alfalfa /%93.48100.0071.7497.8397.83
      Corn-notill /%95.8799.8685.9295.5998.32
      Corn-min /%93.1390.8493.86100.0099.76
      Corn /%92.8394.0992.4199.58100.00
      Grass/pasture /%81.7898.5593.3795.6599.79
      Grass/trees /%98.4998.7798.4999.7399.86
      Grass/pasture-mowed /%89.29100.00100.00100.00100.00
      Hay-windrowed /%99.58100.00100.00100.00100.00
      Oats /%90.0085.0055.0040.0080.00
      Soybeans-notill /%95.8889.5183.5499.5999.38
      Soybeans-min /%98.2183.9597.8499.9699.76
      Soybeans-clean /%96.4698.4893.7697.6499.49
      Wheat /%98.05100.0099.5194.15100.00
      Woods /%99.6099.6098.9799.4599.92
      Building-grass-trees-drives /%97.9398.1997.1599.2298.45
      Stone-steel towers /%97.8598.9282.8096.7796.77
      OA /%96.4393.8693.8998.5999.43
      AA /%94.9095.9990.2794.7098.08
      Kappa coefficient95.9393.0393.0198.4099.35
      Time /min1.895.3832.2718.0712.43
    • Table 5. Classification results of the Pavia University data set

      View table

      Table 5. Classification results of the Pavia University data set

      Classification2D_CNN3D_CNNM3RCNN3D_DenseNetMDSSAN
      Asphalt /%98.9799.4099.2698.6699.82
      Meadows /%99.4399.6599.9899.9299.96
      Gravel /%97.2894.3397.1499.2997.76
      Trees /%99.7197.7899.6198.4799.45
      Sheets /%99.7098.88100.00100.0099.93
      Baresoil /%99.6499.9887.09100.0099.96
      Bitumen /%84.2198.8798.8799.0299.32
      Bricks /%97.9199.3599.57100.0099.78
      Shadows /%91.6699.89100.0097.9999.37
      OA /%98.5399.1898.1299.5499.74
      AA /%96.5098.6897.9599.2699.48
      Kappa coefficient98.0598.9297.4999.3999.66
      Time /min6.617.2173.9276.1019.58
    • Table 6. Classification results of the KSC data set

      View table

      Table 6. Classification results of the KSC data set

      Classification2D_CNN3D_CNNM3RCNN3D_DenseNetMDSSAN
      Scrub /%97.7799.7499.7498.16100.00
      Willow swamp /%78.6099.1893.4288.4895.06
      CP hammock /%87.8993.7598.8398.83100.00
      Slash pine /%70.2458.3386.5188.8995.24
      Oak/broadleaf /%78.8880.1285.0993.1793.79
      Hardwood /%93.8980.3493.0192.14100.00
      Swap /%98.1078.1097.14100.00100.00
      Graminoid marsh /%89.3399.5498.8499.0798.84
      Spartina marsh /%90.7798.65100.0099.8199.62
      Cattail marsh /%90.8499.26100.00100.0098.51
      Salt marsh /%99.76100.0099.5299.76100.00
      Mud flats /%99.4097.0294.6397.2298.81
      Water /%99.89100.0099.89100.00100.00
      OA /%93.0795.1697.4597.6698.98
      AA /%90.4191.0895.8996.5898.45
      Kappa coefficient92.2894.6197.1697.3998.87
      Time /min1.111.7418.0411.258.19
    • Table 7. Effect of three-dimensional convolution module before and after the improvement

      View table

      Table 7. Effect of three-dimensional convolution module before and after the improvement

      Data setModel parameterOA/%
      Before improvementAfter improvementBefore improvementAfter improvement
      Indian Pines118604456516798.7099.43
      Pavia University118474956387299.1799.74
      KSC118548956461295.9398.98
    Tools

    Get Citation

    Copy Citation Text

    Xin Wang, Yanguo Fan. Hyperspectral Image Classification Based on Modified DenseNet and Spatial Spectrum Attention Mechanism[J]. Laser & Optoelectronics Progress, 2022, 59(2): 0210014

    Download Citation

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

    Category: Image Processing

    Received: Jan. 21, 2021

    Accepted: Mar. 15, 2021

    Published Online: Dec. 23, 2021

    The Author Email: Wang Xin (3166588225@qq.com)

    DOI:10.3788/LOP202259.0210014

    Topics