Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 6, 833(2024)

Hyperspectral image classification based on spatial pyramid attention mechanism combined with ResNet

He LIU1, Yingluo SONG2, Longxiang HU1, Guohui LIU1, Kan WANG1, and Aili WANG2、*
Author Affiliations
  • 1Integrated Data Center,State Grid Heilongjiang Electric Power Co. Ltd.,Harbin 150010,China
  • 2Heilongjiang Province Key Laboratory of Laser Spectroscopy Technology and Application,School of Measurement-Control Technology and Communications Engineering,Harbin University of Science and Technology,Harbin 150080,China
  • show less
    Figures & Tables(16)
    HSI classification model architecture of DSPCRN
    Structure of SPCS
    Structure of SPCE
    Structure of dynamic convolution
    Mish activation function
    MUUFL dataset
    Trento dataset
    Classification results on MUUFL dataset
    Classification results on Trento dataset
    Loss and accuracy curves of MUUFL dataset
    • Table 1. MUFFL dataset

      View table
      View in Article

      Table 1. MUFFL dataset

      序号对应颜色类别名样本数
      1树木23 246
      2大片草地4 270
      3混合地面6 882
      4泥土和沙子1 826
      5道路6 687
      6466
      7建筑阴影2 233
      8建筑6 240
      9人行道1 385
      10黄色人行道边183
      11环保板269
      共计53 267
    • Table 2. Trento dataset

      View table
      View in Article

      Table 2. Trento dataset

      序号对应颜色类别名样本数
      共计53 267
      1苹果树4 034
      2建筑物2 903
      3地面479
      4树木9 123
      5橄榄园10 591
      6道路3 174
    • Table 3. Comparison of classification accuracy of MUUFL dataset

      View table
      View in Article

      Table 3. Comparison of classification accuracy of MUUFL dataset

      类别RBF-SVMEMP-SVMCNNResNetPyResNetSSRNOurs
      189.89±3.9693.23±2.1793.01±3.8597.11±0.8697.87±0.4696.61±0.2597.32±0.13
      270.20±3.6268.08±2.9975.12±6.1392.67±2.5692.01±0.2791.28±1.7591.25±0.07
      365.44±7.8276.76±1.6379.65±6.1276.69±4.3683.35±0.3684.48±0.8890.01±1.45
      479.35±7.6888.26±4.2775.08±9.6391.96±0.2192.33±0.7892.86±0.6591.78±2.01
      584.44±2.5285.08±2.8390.15±1.0890.69±1.0292.89±0.9390.97±0.5693.03±1.02
      697.95±2.2381.88±1.5353.09±9.1186.99±3.6087.32±0.1394.33±0.9695.41±1.16
      764.38±5.2169.45±0.5366.38±1.2392.18±3.9091.21±1.2292.29±0.1292.58±3.83
      890.30±1.4393.85±2.0594.12±1.5194.82±0.4895.87±1.0194.74±0.4996.72±0.53
      953.84±1.5146.78±20.5067.64±0.0665.40±0.3472.57±2.1071.77±5.2272.22±3.22
      1074.04±9.7676.61±2.1292.46±0.46100.0±0.0087.64±0.2372.22±2.2273.79±5.91
      1154.32±0.8244.56±1.6294.91±0.4991.73±0.9194.44±0.0190.62±0.5093.96±0.28
      OA/%81.90±2.0083.73±0.0286.94±1.4391.36±0.6792.17±1.1892.81±1.7494.08±0.13
      AA/%74.92±4.2374.96±3.8480.11±3.5889.11±1.7088.38±0.6888.38±2.8089.82±1.78
      100K75.64±3.0678.54±2.8282.59±2.0988.57±3.0189.22±0.4190.45±0.8091.17±0.17
      Time/s65,34330.05267.24335.32388.67297.64280.22
    • Table 4. Comparison of classification accuracy of Trento dataset

      View table
      View in Article

      Table 4. Comparison of classification accuracy of Trento dataset

      类别RBF-SVMEMP-SVMCNNResNetPyResNetSSRNOurs
      182.59±8.5394.38±4.2696.53±3.0294.85±2.1699.31±0.4597.60±2.7698.90±0.06
      283.56±3.9284.63±1.0182.95±0.6883.52±6.4188.67±0.2188.45±6.9589.50±5.01
      396.89±2.2095.33±0.0297.61±3.0392.91±1.0094.91±0.7294.91±0.7298.48±1.22
      495.37±2.1095.36±1.0689.01±0.0899.48±0.4596.84±1.6499.15±1.1799.98±0.01
      592.40±2.7696.48±0.0994.92±0.3599.25±0.0198.88±0.0998.45±1.7898.68±0.13
      677.43±4.6175.17±2.3587.41±6.3781.47±0.9383.95±0.0188.52±0.7195.99±1.73
      OA/%89.52±1.2591.72±1.2593.17±1.2794.84±0.0995.10±1.8096.19±0.5698.32±0.41
      AA/%88.04±0.7090.22±1.4691.37±2.2591.91±1.0193.76±0.5294.51±0.8996.92±1.36
      100K85.91±1.7289.33±0.8990.80±1.7393.11±1.2593.77±0.8994.91±0.7696.29±0.35
      Time/s157.26424.33294.82240.19278.11220.13210.46
    • Table 5. Comparison of ablation experiments

      View table
      View in Article

      Table 5. Comparison of ablation experiments

      数据集指标SSRNSSRN +DCDC+SPCA+SSRN
      MUUFLOA/%92.81±1.7493.48±0.3894.08±0.13
      AA/%88.38±2.8088.31±0.8189.82±1.78
      100K90.45±0.8091.45±0.4492.17±0.17
      TrentoOA/%96.19±0.5697.46±0.9598.32±0.41
      AA/%94.51±0.8996.80±0.2896.92±1.36
      100K94.91±0.7695.17±0.0296.29±0.35
    • Table 6. Comparison of SPC module’s connection times

      View table
      View in Article

      Table 6. Comparison of SPC module’s connection times

      数据集指标×1×2×3
      MUUFLOA/%92.78±0.6993.62±0.8794.08±0.13
      AA/%88.19±2.0988.12±0.4289.82±1.78
      100K90.43±0.0995.11±0.1792.17±0.17
      TrentoOA/%97.05±0.4597.89±0.9598.32±0.41
      AA/%94.27±1.3695.36±0.2896.92±1.36
      100K95.22±0.1195.84±0.3996.29±0.35
    Tools

    Get Citation

    Copy Citation Text

    He LIU, Yingluo SONG, Longxiang HU, Guohui LIU, Kan WANG, Aili WANG. Hyperspectral image classification based on spatial pyramid attention mechanism combined with ResNet[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(6): 833

    Download Citation

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

    Category: Research Articles

    Received: May. 10, 2023

    Accepted: --

    Published Online: Jul. 30, 2024

    The Author Email: Aili WANG (aili925@hrbust.edu.cn)

    DOI:10.37188/CJLCD.2023-0175

    Topics