Acta Optica Sinica, Volume. 40, Issue 2, 0228001(2020)

Spatially-Regularized Manifold Discriminant Analysis Algorithm for Hyperspectral Image Classification

Hong Huang*, Lihua Wang, and Guangyao Shi
Author Affiliations
  • Key Laboratory of Optoelectronic Technique System of the Ministry of Education, Chongqing University, Chongqing 400044, China
  • show less
    Figures & Tables(12)
    Flow chart of SSRMDA algorithm
    ERS segmentation and ground-truth images. (a) ERS segmentation image; (b) ground-truth image
    Hyperspectral images in Indian Pines dataset. (a) False-color image; (b) ground-truth image
    Hyperspectral images in Washington DC Mall dataset. (a) False-color image; (b) ground-truth image
    Overall classification accuracy of SSRMDA algorithm with different values of K and Kb on different datasets. (a) Indian Pines dataset; (b) Washington DC Mall dataset
    Overall classification accuracy of SSRMDA algorithm with different α values
    Classification diagrams of different algorithms on Indian Pines dataset
    Classification diagrams of different algorithms on Washington DC Mall dataset
    • Table 1. Classification accuracy of different algorithms on Indian Pines dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation of ov

      View table

      Table 1. Classification accuracy of different algorithms on Indian Pines dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation of ov

      Algorithmni=5ni=10ni=15ni=20ni=30
      RAW51.81±2.37(0.463)59.48±1.72(0.547)65.43±1.39(0.613)68.45±1.03(0.645)71.72±1.04(0.681)
      PCA51.73±2.39(0.462)59.30±1.71(0.545)65.17±1.36(0.610)68.22±0.98(0.643)71.42±1.05(0.678)
      NPE50.17±2.32(0.445)56.65±1.81(0.516)61.86±1.67(0.574)64.78±1.35(0.605)67.02±1.24(0.629)
      LPP52.08±2.09(0.467)59.47±1.80(0.547)64.87±1.68(0.607)67.13±1.35(0.631)70.10±1.08(0.663)
      LDA54.82±2.38(0.498)63.71±1.70(0.595)69.87±1.55(0.662)72.26±1.06(0.688)75.12±1.10(0.719)
      MFA62.10±3.79(0.577)74.82±2.03(0.716)80.05±1.35(0.775)82.99±1.20(0.807)86.51±1.09(0.846)
      LGSFA60.80±4.10(0.563)72.56±1.94(0.691)80.38±1.62(0.778)83.71±0.99(0.815)88.05±1.20(0.864)
      DSSM54.15±3.30(0.490)62.63±2.95(0.582)72.08±1.51(0.686)74.44±1.23(0.712)76.91±1.17(0.739)
      LPNPE56.81±3.37(0.522)70.14±2.27(0.666)76.89±1.60(0.740)81.97±1.55(0.796)88.26±0.94(0.866)
      SSRLDE56.85±3.19(0.523)71.38±2.80(0.680)78.97±2.64(0.763)83.34±1.97(0.812)89.46±0.99(0.880)
      SSRMDA70.80±2.87(0.672)82.14±2.65(0.798)86.22±1.82(0.844)89.21±1.51(0.876)91.58±1.34(0.904)
    • Table 2. Classification accuracy of each-class grand object on Indian Pines dataset obtained by different algorithms%

      View table

      Table 2. Classification accuracy of each-class grand object on Indian Pines dataset obtained by different algorithms%

      ClassRAWPCANPELPPLDAMFALGSFADSSMLPNPESSRLDESSRMDA
      Alfalfa97.2297.2297.2297.2210010010097.22100100100
      Corn-Notill61.3361.1155.8260.2569.2683.5682.6359.1176.5579.6386.92
      Corn-mintill58.5558.5552.8954.7367.5275.8976.9956.5876.2677.6189.54
      Corn53.3053.7443.6148.0167.8492.0793.8344.0585.9090.3188.99
      Grass-pasture71.4671.2463.6367.4483.5189.2189.4373.5789.2189.8591.12
      Grass-trees90.4990.3585.1782.5197.9099.4499.5893.9882.9379.8699.58
      Grassp-asture-mowed100100100100100100100100100100100
      Hay-windrowed87.8287.8277.3581.1971.5897.4397.6588.0392.5298.5097.43
      Oats100100100100100100100100100100100
      Soybean-nottill77.5476.8169.0470.6275.3489.0885.4180.0684.8986.4686.25
      Soybean-mintill74.9374.7368.7070.1573.7383.4185.8272.1981.2183.4192.02
      Soybean-clean47.3346.6443.7146.1246.4786.2378.3143.3771.7783.82190.88
      Wheat96.9296.9296.9297.9497.9498.9798.9796.9298.4698.9798.97
      Woods76.5376.2173.9575.5678.3082.9084.2776.5384.1985.8085.08
      Buildings-Grass-Trees-Drives71.0171.2768.8869.1576.5985.6384.8472.8791.4988.0386.97
      Stone-Steel-Towers10010097.5910010010010098.79100100100
      OA72.2472.0166.6168.5474.6686.3786.3171.3382.4684.6690.37
      AA79.0378.9174.6576.3181.6291.4991.1178.3388.4690.1493.36
      Kappa0.6840.6820.6200.6420.7120.8450.8440.6740.8000.8260.890
    • Table 3. Classification accuracy of different algorithms on Washington DC Mall dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation

      View table

      Table 3. Classification accuracy of different algorithms on Washington DC Mall dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation

      Algorithmni=5ni=10ni=15ni=20ni=30
      RAW80.86±2.96(0.761)82.80±2.23(0.786)84.07±2.17(0.802)85.87±1.86(0.824)86.60±1.22(0.833)
      PCA80.86±2.96(0.761)82.79±2.23(0.786)84.06±2.18(0.802)85.86±1.86(0.824)86.58±1.23(0.833)
      NPE79.91±3.33(0.750)81.20±3.06(0.767)82.91±2.13(0.788)84.94±1.74(0.812)85.28±1.67(0.817)
      LPP80.21±3.35(0.753)82.24±1.91(0.779)83.47±1.87(0.795)85.46±1.81(0.819)86.39±1.72(0.831)
      LDA81.56±3.33(0.770)83.08±2.23(0.790)85.52±1.77(0.820)86.69±1.48(0.834)87.84±1.27(0.848)
      MFA84.49±2.51(0.806)87.92±2.25(0.849)90.98±2.20(0.887)91.87±0.89(0.898)92.70±0.59(0.908)
      LGSFA85.48±2.68(0.818)88.53±2.20(0.857)91.28±1.15(0.891)92.47±1.03(0.906)93.64±0.98(0.920)
      DSSM80.63±3.31(0.759)82.80±2.19(0.786)84.08±2.57(0.802)85.33±2.11(0.817)86.61±1.22(0.833)
      LPNPE78.71±5.15(0.736)86.65±2.63(0.834)87.75±2.36(0.847)89.14±1.02(0.864)90.34±1.06(0.879)
      SSRLDE79.45±5.36(0.744)86.51±4.22(0.832)87.34±3.30(0.843)89.62±1.37(0.870)90.86±1.19(0.886)
      SSRMDA88.73±2.15(0.859)92.78±1.19(0.910)94.04±0.98(0.925)95.35±0.73(0.942)96.67±0.73(0.958)
    • Table 4. Classification accuracy of each-class ground object on Washington DC Mall dataset obtained by different algorithms%

      View table

      Table 4. Classification accuracy of each-class ground object on Washington DC Mall dataset obtained by different algorithms%

      ClassRAWPCANPELPPLDAMFALGSFADSSMLPNPESSRLDESSRMDA
      Road95.0495.0194.7194.3894.7697.8398.7195.1793.7295.0699.79
      Water94.1294.1293.5794.7496.3397.2498.2294.1399.2696.4598.39
      Building87.6787.6486.5686.5388.3391.4093.1587.6496.9395.3998.61
      Vegetation97.3497.3497.0197.2797.3797.8897.8897.3497.0896.4098.73
      Trail66.8366.8964.0469.3972.3589.5293.2066.6691.8682.3893.76
      Shadow67.9067.9065.0367.7368.2073.7880.4667.8672.8572.0582.66
      OA88.0988.0887.0587.9188.8792.8194.6488.1192.9091.6496.70
      AA84.8284.8283.4985.0186.2291.2893.6084.8091.9589.6295.49
      Kappa0.8500.8490.8370.8480.8600.9090.9330.8500.9110.8950.959
    Tools

    Get Citation

    Copy Citation Text

    Hong Huang, Lihua Wang, Guangyao Shi. Spatially-Regularized Manifold Discriminant Analysis Algorithm for Hyperspectral Image Classification[J]. Acta Optica Sinica, 2020, 40(2): 0228001

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: Jul. 18, 2019

    Accepted: Sep. 6, 2019

    Published Online: Jan. 2, 2020

    The Author Email: Huang Hong (hhuang@cqu.edu.cn)

    DOI:10.3788/AOS202040.0228001

    Topics