• Journal of Geo-information Science
  • Vol. 22, Issue 10, 2010 (2020)
Hongshu HE1、2, Xiaoxia HUANG1、*, Hongga LI1, Lingjia NI1、2, Xinge WANG3, Chong CHEN3, and Ze LIU3
Author Affiliations
  • 1Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
  • 2University of Chinese Academy of Sciences, Beijing 100049, China
  • 3Urban and Rural Planning Management Center of the Ministry of Housing and Urban-Rural Development,Beijing 100835, China
  • show less
    DOI: 10.12082/dqxxkx.2020.190622 Cite this Article
    Hongshu HE, Xiaoxia HUANG, Hongga LI, Lingjia NI, Xinge WANG, Chong CHEN, Ze LIU. Water Body Extraction of High Resolution Remote Sensing Image based on Improved U-Net Network[J]. Journal of Geo-information Science, 2020, 22(10): 2010 Copy Citation Text show less
    Technical routes for extracting the water body
    Fig. 1. Technical routes for extracting the water body
    U-Net architecture
    Fig. 2. U-Net architecture
    Improved U-Net network low-dimensional information enhancement
    Fig. 3. Improved U-Net network low-dimensional information enhancement
    Full connection condition random field post-processing model
    Fig. 4. Full connection condition random field post-processing model
    Location of Qingdao study area
    Fig. 5. Location of Qingdao study area
    GF-2 image processing flowchart
    Fig. 6. GF-2 image processing flowchart
    Training curve of improved U-Net
    Fig. 7. Training curve of improved U-Net
    感受野步长填充输出大小
    InputRGBimage:3@256×256
    Conv+ReLU3×31164@256×256
    Conv+ReLU3×31164@256×256
    Max-pooling64@128×128
    Conv+ReLU3×311128@128×128
    Conv+ReLU3×311128@128×128
    Max-pooling128@64×64
    Conv+ReLU3×311256@64×64
    Conv+ReLU3×311256@64×64
    Conv+ReLU3×311256@64×64
    Max-pooling256@32×32
    Conv+ReLU3×311512@32×32
    Conv+ReLU3×311512@32×32
    Conv+ReLU3×311512@32×32
    Max-pooling512@16×16
    Conv+ReLU3×311512@16×16
    Conv+ReLU3×311512@16×16
    Conv+ReLU3×311512@16×16
    Max-pooling512@8×8
    Table 1. VGG16 network structure configuration
    影像编号中心经度/°E中心纬度/°N成像时间影像大小/像素×像素
    L1A0003593712120.536.72018-11-1227 620×35 273
    L1A0003593719120.436.32018-11-1227 620×35 113
    L1A0003593868120.636.32018-11-1227 620×35 191
    Table 2. RemotesensingimageinformationintheQingdaostudyarea
    项目系统CPU内存硬盘显卡
    内容Ubuntu16.04Intel E5-16308 GB500 GBNVIDIA GTX970
    Table 3. Basic system platform configuration
    项目GPU-DriverCUDAPythonKerasTensorflow-gpu
    内容3848.03.62.2.41.4.0
    Table 4. Important software configuration
    Table 5. Comparison of water extraction results by different methods in 5 typical areas of the study area
    实际正类实际负类
    预测正类TPFP
    预测负类FNTN
    Table 6. Confusion matrix for accuracy evaluation
    方法IoU/%精准率/%Kappa系数
    SegNet77.682.50.79
    经典U-net82.390.40.88
    改进后的U-Net网络88.194.80.93
    Table 7. Accuracy comparison of water extraction results
    影像编号中心经度/°E中心纬度/°N成像时间影像大小/像素×像素
    L1A0003553729120.136.32018-10-28276 20×292 00
    L1A0003351642101.536.82018-07-26276 20×292 00
    Table 8. Remote sensing image information in the application area
    区域1区域2区域3区域4区域5
    青岛原始影像
    水体信息
    西宁原始影像
    水体信息
    Table 9. Comparison of water extraction results in 5 typical areas of the application area
    Hongshu HE, Xiaoxia HUANG, Hongga LI, Lingjia NI, Xinge WANG, Chong CHEN, Ze LIU. Water Body Extraction of High Resolution Remote Sensing Image based on Improved U-Net Network[J]. Journal of Geo-information Science, 2020, 22(10): 2010
    Download Citation