• Laser & Optoelectronics Progress
  • Vol. 58, Issue 14, 1401002 (2021)
Haitao Wang1, Yichen Wang1, Yongqiang Wang2, and Yurong Qian1、*
Author Affiliations
  • 1College of Software, Xinjiang University, Urumqi, Xinjiang 830046, China
  • 2College of Information Engineering and Science, Xinjiang University, Urumqi, Xinjiang 830046, China
  • show less
    DOI: 10.3788/LOP202158.1401002 Cite this Article Set citation alerts
    Haitao Wang, Yichen Wang, Yongqiang Wang, Yurong Qian. Cloud Detection of Landsat Image Based on MS-UNet[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1401002 Copy Citation Text show less
    UNet network structure
    Fig. 1. UNet network structure
    UNet feature extraction module
    Fig. 2. UNet feature extraction module
    Multi-scale feature extraction module
    Fig. 3. Multi-scale feature extraction module
    Schematic diagrams of activation function. (a) ReLU function: max(x,0); (b) FReLU function: max[x,T(x)]
    Fig. 4. Schematic diagrams of activation function. (a) ReLU function: max(x,0); (b) FReLU function: max[x,T(x)]
    Flowchart of improved UNet
    Fig. 5. Flowchart of improved UNet
    Flowchart of improved UNet
    Fig. 6. Flowchart of improved UNet
    Curves of train loss
    Fig. 7. Curves of train loss
    Comparison of results of improved modules. (a) (b) Broken clouds; (c) (d) thin clouds
    Fig. 8. Comparison of results of improved modules. (a) (b) Broken clouds; (c) (d) thin clouds
    Comparison of cloud detection results of different methods. (a) (b) Broken clouds; (c) (d) thin clouds
    Fig. 9. Comparison of cloud detection results of different methods. (a) (b) Broken clouds; (c) (d) thin clouds
    MethodPrecisionAccuracyRecallF1-ScoreIoUMIoU
    UNet0.8900.8680.7870.8350.7170.760
    MS-UNet0.8810.9360.9410.9120.8390.871
    MS-UNet+FReLU0.9010.9430.9450.9210.8550.884
    Table 1. Index comparison results between improved algorithm and original algorithms
    MethodPrecisionAccuracyRecallF1-ScoreIoUMIoU
    MF-CNN[19]0.8930.8780.8860.9140.7950.833
    SegNet[20]0.8280.8730.8330.8310.7110.763
    DeepLabV3_ResNet50[21]0.9310.9300.8770.9090.8330.863
    DeepLabV3_ResNet101[21]0.9120.9380.9230.9180.8480.877
    MS-UNet0.8810.9360.9410.9120.8390.871
    MS-UNet+FReLU0.9010.9430.9450.9210.8550.884
    Table 2. Comparison of indicators of different algorithms
    Haitao Wang, Yichen Wang, Yongqiang Wang, Yurong Qian. Cloud Detection of Landsat Image Based on MS-UNet[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1401002
    Download Citation