Optics and Precision Engineering, Volume. 32, Issue 22, 3348(2024)

Spatial-spectral reweighted sparse multi-layer nonnegative matrix factorization for hyperspectral image unmixing

Jiming TANG1... Wenxing BAO1,*, Bingbing LEI1,*, Wei FENG2 and Kewen QU1 |Show fewer author(s)
Author Affiliations
  • 1School of Computer Science and Engineering, North Minzu University, Yinchuan75002, China
  • 2School of Electronic Engineering, Xidian University, Xi'an710071, China
  • show less
    Figures & Tables(18)
    Flowchart of the SSRS-MLNMF
    Experimental results of different models dealing with synthetic datasets
    First row (a) shows the true abundance and the second row (b) shows the abundance obtained by the model
    Comparison of the estimated endmember curves with the true endmember curves for the synthetic dataset
    Comparison between the abundance obtained by the model and the true abundance
    Comparison of the true endmember curves with the reference endmember curve
    Abundance map obtained by the model (From top to bottom, the first row:Alunite, Pyrope, Muscovite, Andradite .The second row: Dumortierite, Montmorillonite, Sphene, Kaolinite_2 . The third row: Nontronite, Chalcedony, Buddingtonite, Kaolinite_1)
    Comparison of the true endmember curves with the reference endmember curve (From left to right, from top to bottom, the first row of subgraphs are: Alunite, Andradite, Buddingtonite, Dumortierite. The second row of subgraphs are:Kaolinite_1, Kaolinite_2, Muscovite, Montmorillonite.The third row of subgraphs are: Nontronite, Pyrope, Sphene, Chalcedony.)
    • Table 1. [in Chinese]

      View table
      View in Article

      Table 1. [in Chinese]

      算法1:SSRS-MLNMF算法

      输入:高光谱图像数据XRL×N,端元数量P,参数α0,τ,Lmax,Tmax

      输出:端元矩阵ERL×P和丰度矩阵SRP×N

      初始化:A,S,计算空间权重和光谱权重并融合得到Wspa-spe,计算高光谱粗略图,并构建去噪权重V

      Repeat:

       1. Al+1Al.*XlSlT./AlSlSlT+14αAAl-34

       2. Sl+1Sl.*AlTXl./AlTAlSl+14αSVVSl-34

       3.更新权重因子:Vt+1=1St2,1+ξ

       4.更新条件:k+1k

      停止条件:直到满足停止条件。

      结束:返回结果:A=l=1LAl, S=SL

    • Table 1. mSAD comparison of different algorithms under different noise levels (synthetic dataset)

      View table
      View in Article

      Table 1. mSAD comparison of different algorithms under different noise levels (synthetic dataset)

      SNR/算法VCAL1-NMFMLNMFDMFPro-BM3DProposed
      SNR=10 dB0.251 20.118 50.105 40.226 90.036 50.041 3
      SNR=20 dB0.207 60.070 10.068 50.207 30.041 40.039 9
      SNR=30 dB0.183 90.064 80.064 30.126 40.040 50.038 7
      SNR=40 dB0.119 30.050 50.059 30.106 60.036 10.035 8
    • Table 2. RMSE comparison of different algorithms under different noise levels (synthetic dataset)

      View table
      View in Article

      Table 2. RMSE comparison of different algorithms under different noise levels (synthetic dataset)

      SNR/算法VCAL1-NMFMLNMFDMFPro-BM3DProposed
      SNR=10 dB0.137 901 4210.057 90.142 10.060 20.053 9
      SNR=20 dB0.093 80.107 40.055 90.121 90.049 80.047 2
      SNR=30 dB0.105 80.096 50.048 20.101 50.046 40.045 1
      SNR=40 dB0.102 80.082 40.046 40.081 50.047 20.043 7
    • Table 3. Optimal structural parameters of the model (synthetic dataset)

      View table
      View in Article

      Table 3. Optimal structural parameters of the model (synthetic dataset)

      层数每层迭代次数mSAD
      51 0000.076 9
      58000.060 2
      101 0000.029 1
      108000.051 2
      158000.048 7
      155000.057 1
    • Table 4. Performance of different algorithms on synthetic dataset

      View table
      View in Article

      Table 4. Performance of different algorithms on synthetic dataset

      VCAL1-NMFMLNMFDMFPro-BM3DProposed
      Almandine0.253 40.264 90.034 40.170 40.032 00.017 1
      Ammonio0.583 70.333 60.037 30.051 80.016 30.019 0
      Antigorite0.153 70.130 80.104 80.200 20.038 30.025 8
      Axinite0.639 10.122 90.123 70.176 90.046 00.061 3
      Biotite0.191 90.275 90.113 20.289 80.060 40.025 8
      Carnallite0.349 30.397 60.027 80.139 30.023 30.025 4
      mSAD0.361 90.254 20.073 50.171 40.036 10.029 1
      mRMSE0.087 80.086 70.064 10.041 50.059 60.038 1
    • Table 5. Optimal structural parameters of the model (Jasper Ridge dataset)

      View table
      View in Article

      Table 5. Optimal structural parameters of the model (Jasper Ridge dataset)

      层数每层迭代次数mSAD
      406000.083 1
      408000.088 2
      506000.051 6
      508000.056 4
      606000.066 9
      608000.087 3
    • Table 6. Performance of different algorithms on the Jasper Ridge dataset

      View table
      View in Article

      Table 6. Performance of different algorithms on the Jasper Ridge dataset

      VCAL1-NMFMLNMFDMFPro-BM3DProposed
      Tree0.197 90.105 60.190 30.050 70.053 20.047 1
      Water0.445 50.196 40.150 00.139 10.103 50.106 8
      Dirt0.279 10.362 10.238 70.206 00.025 60.030 2
      Road0.265 60.402 90.196 60.116 50.031 10.022 4
      mSAD0.297 00.266 70.193 90.128 10.053 40.051 6
      mRMSE0.197 90.105 60.090 30.050 70.053 20.046 3
    • Table 7. Optimal structural parameters of the model (Cuprite dataset)

      View table
      View in Article

      Table 7. Optimal structural parameters of the model (Cuprite dataset)

      层数每层迭代次数mSAD
      151 0000.142 8
      151 5000.153 7
      201 0000.144 0
      201 5000.145 5
      251 0000.138 6
      251 5000.110 9
    • Table 8. Performance of different algorithms on the Cuprite dataset

      View table
      View in Article

      Table 8. Performance of different algorithms on the Cuprite dataset

      VCAL1NMFMLNMFDMFPro-BM3DProposed
      #1 Alunite1.141 00.052 50.112 61.0780..1 8650.177 4
      #2 Andradite0.126 70.075 80.087 60.086 60.154 30.081 1
      #3 Buddingtonite0.199 80.078 10.134 50.143 70.107 60.175 5
      #4 Dumortierite0.277 90.083 40.115 70.951 10.117 10.133 3
      #5 Kaolinite_10.158 50.083 70.085 30.115 70.091 40.082 0
      #6 Kaolinite_20.101 10.094 30.058 10.068 90.224 80.149 5
      #7 Muscovite0.174 80.102 11.363 61.003 50.094 00.024 6
      #8 Montmorillonite0.108 50.102 30.072 40.057 90.080 00.130 2
      #9 Nontronite0.131 20.112 10.190 60.092 80.169 80.111 4
      #10 Pyrope0.141 20.146 20.089 60.112 90.110 80.089 3
      #11Sphene0.264 20.146 90.164 70.101 40.157 80.075 6
      #12Chalcedony0.133 90.174 00.1361.017 60.161 50.101 7
      mSAD0.246 60.116 40.217 60.402 50.138 00.110 9
    • Table 9. Sum the time taken by each algorithm to process different datasets

      View table
      View in Article

      Table 9. Sum the time taken by each algorithm to process different datasets

      VCAL1NMFMLNMFDMFPro-BM3DProposed
      Synthetic dataset0.84.213.68.620.313.6
      Samson1.220.451.310.4140.7102.6
      Jasper Ridge1.918.355.415.2162.7143.6
      Urban2.323.867.218.91 258.8984.6
      Cuprite4.928.277.325.82 189.21 612.4
    Tools

    Get Citation

    Copy Citation Text

    Jiming TANG, Wenxing BAO, Bingbing LEI, Wei FENG, Kewen QU. Spatial-spectral reweighted sparse multi-layer nonnegative matrix factorization for hyperspectral image unmixing[J]. Optics and Precision Engineering, 2024, 32(22): 3348

    Download Citation

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

    Category:

    Received: Jun. 11, 2024

    Accepted: --

    Published Online: Mar. 10, 2025

    The Author Email: BAO Wenxing (bwx71@163. com), LEI Bingbing (x_generation@126.com)

    DOI:10.37188/OPE.20243222.3348

    Topics