• Laser & Optoelectronics Progress
  • Vol. 56, Issue 3, 031003 (2019)
Min Wang, Tanfei Fan*, Weiguo Yun, and Zhihui Wang
Author Affiliations
  • School of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an, Shaanxi 710055, China
  • show less
    DOI: 10.3788/LOP56.031003 Cite this Article Set citation alerts
    Min Wang, Tanfei Fan, Weiguo Yun, Zhihui Wang. PFWG Improved CNN Multispectra Remote Sensing Image Classification[J]. Laser & Optoelectronics Progress, 2019, 56(3): 031003 Copy Citation Text show less
    Schematic of CNN
    Fig. 1. Schematic of CNN
    Research area image. (a) Study area 1; (b) study area 2
    Fig. 2. Research area image. (a) Study area 1; (b) study area 2
    Cross-track illumination radiation correction diagram
    Fig. 3. Cross-track illumination radiation correction diagram
    Radiation correction diagram
    Fig. 4. Radiation correction diagram
    MNF result graph
    Fig. 5. MNF result graph
    Diagram of MNF
    Fig. 6. Diagram of MNF
    BP neural network classification result graph
    Fig. 7. BP neural network classification result graph
    CNN classification result graph
    Fig. 8. CNN classification result graph
    PFWG improved CNN classification result graph
    Fig. 9. PFWG improved CNN classification result graph
    Classification result comparison graph. (a) BP neural network classification result; (b) CNN classification result; (c) PFWG improved CNN classification result
    Fig. 10. Classification result comparison graph. (a) BP neural network classification result; (b) CNN classification result; (c) PFWG improved CNN classification result
    ImageFeature typeImage characteristic
    BuildingBuildings are arranged neatly and there are roads passing through.The colors are gray and dark gray.
    WaterUniform texture, smooth borders, blue-green or dark blue.
    Green areaBlocky distribution, the color characteristics are more obvious, dark green, light blue.
    NudationTexture shows a pattern, there are roads through, the color is purple, gray purple.
    Table 1. Land cover classification system and interpretation marks in the study area
    ClassificationBuildingNudationWaterGreenbeltAggregate
    Building91.4419.736.073.4730.1
    Nudation8.5680.270022.1
    Water0093.93023.3
    Greenbelt00096.5324.5
    Aggregate100100100100100
    Table 2. BP neural network classification result confusion matrix%
    ClassificationBuildingNudationWaterGreenbeltAggregate
    Building93.1217.852.940.7128.6
    Nudation6.8882.150022.2
    Water0097.06024.2
    Greenbelt00099.2925.0
    Aggregate100100100100100
    Table 3. CNN classification result confusion matrix%
    ClassificationBuildingNudationWaterGreenbeltAggregate
    Building96.6315.480.82028.3
    Nudation3.3784.520021.9
    Water0099.18024.8
    Greenbelt00010025.0
    Aggregate100100100100100
    Table 4. PFWG improved CNN classification result confusion matrix%
    ParameterBP neural networkConvolutional neural networkImproved algorithm
    Kappa coefficient0.880.900.94
    Classification speed /min2574
    Overall accuracy /%91.6392.8293.73
    Table 5. Classification accuracy evaluation matrix
    Min Wang, Tanfei Fan, Weiguo Yun, Zhihui Wang. PFWG Improved CNN Multispectra Remote Sensing Image Classification[J]. Laser & Optoelectronics Progress, 2019, 56(3): 031003
    Download Citation