• Acta Photonica Sinica
  • Vol. 51, Issue 2, 0210007 (2022)
Tianwei YU1, Enrang ZHENG1、*, Junge SHEN2, and Kai WANG3
Author Affiliations
  • 1School of Electrical and Control Engineering,Shaanxi University of Science and Technology,Xi′an710021,China
  • 2Unmanned System Research Institute,Northwestern Polytechnical University,Xi′an710072,China
  • 3Henan Key Laboratory of Underwater Intelligent Equipment,Zhengzhou 450000,China
  • show less
    DOI: 10.3788/gzxb20225102.0210007 Cite this Article
    Tianwei YU, Enrang ZHENG, Junge SHEN, Kai WANG. Optical Remote Sensing Image Scene Classification Based on Multi-level Cross-layer Bilinear Fusion[J]. Acta Photonica Sinica, 2022, 51(2): 0210007 Copy Citation Text show less
    Overall flowchart of the proposed method
    Fig. 1. Overall flowchart of the proposed method
    Dilated convolution with different dilation rate
    Fig. 2. Dilated convolution with different dilation rate
    Multiscale dilated convolution module
    Fig. 3. Multiscale dilated convolution module
    Multi-level attention feature fusion module
    Fig. 4. Multi-level attention feature fusion module
    Spatial attention module
    Fig. 5. Spatial attention module
    CM on UCM dataset with a training ratio of 80%
    Fig. 6. CM on UCM dataset with a training ratio of 80%
    Typical samples of forest and parking lot categories and misclassification samples
    Fig. 7. Typical samples of forest and parking lot categories and misclassification samples
    CM on AID dataset with a training ratio of 50%
    Fig. 8. CM on AID dataset with a training ratio of 50%
    Samples from river、port and bridge scene categories of AID dataset
    Fig. 9. Samples from river、port and bridge scene categories of AID dataset
    CM on PatternNet dataset with a training ratio of 20%
    Fig. 10. CM on PatternNet dataset with a training ratio of 20%
    DatasetsImages per classTotal imagesImages sizeScene classesSpatial resolution/m
    UCM1002 100256×256210.3
    AID220~42010 000600×600301~8
    PatternNet80030 400256×256380.062~0.493
    Table 1. Scene dataset information
    dOA(20%)
    12893.02
    25693.08
    51293.11
    1 02493.70
    2 04893.23
    Table 2. Influence of parameter d on classification performance of AID dataset
    ArchitectureMethodsOA(20%)
    1Without MDC module93.28±0.33
    2Without MAFF module91.23±0.25
    3Without CBF92.91±0.22
    4ResNet50+MDC+MAFF+CBF(Ours)93.70±0.11
    Table 3. Ablation experiment on AID dataset with a training ratio of 20%
    MethodsOA(80%)OA(50%)
    GBNet1698.57±0.4897.05±0.19
    ARCNet1799.12±0.4096.81±0.14
    Fusion by Addition1894.72±1.79
    Fine-tuned ResNet501991.9089.43
    Siamese ResNet501994.2990.95
    Two-stream fusion2098.02±1.0396.97±0.75
    CNN-CapsNet1499.05±0.2497.59±0.16
    Ours99.32±0.2098.13±0.18
    Table 4. Classification result comparison on UCM dataset
    MethodsOA(50%)OA(20%)
    ResNet502191.31±0.5888.23±0.70
    GBNet1694.58±0.1292.20±0.23
    ARCNet1793.10±0.5588.75±0.40
    Two-stream fusion2094.58±0.2592.32±0.41
    Ours95.84±0.2693.70±0.11
    Table 5. Classification result comparison on AID dataset
    MethodsOA(50%)OA(20%)
    GLANet(SVM)2299.40±0.2198.91±0.19
    SDAResNet2399.58±0.1099.30±0.08
    VGGNet(SVM)2497.5±0.02
    ResNet101(SVM)2498.6±0.02
    Inception-V3(SVM)2497±0.02
    Ours99.60±0.0699.42±0.05
    Table 6. Classification result comparison on PatternNet dataset
    Tianwei YU, Enrang ZHENG, Junge SHEN, Kai WANG. Optical Remote Sensing Image Scene Classification Based on Multi-level Cross-layer Bilinear Fusion[J]. Acta Photonica Sinica, 2022, 51(2): 0210007
    Download Citation