Acta Optica Sinica, Volume. 39, Issue 12, 1228002(2019)

Remote Sensing Building Detection Based on Binarized Semantic Segmentation

Tianyou Zhu1,2,3, Feng Dong1,2, and Huixing Gong1,2、*
Author Affiliations
  • 1Key Laboratory of Infrared System Detection and Imaging, Chinese Academy of Sciences, Shanghai 200083, China
  • 2Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China
  • 3University of Chinese Academy of Sciences, Beijing 100049, China
  • show less
    Figures & Tables(16)
    Flow chart for training of traditional neural network
    Flow chart for training of binarized neural network
    Comparison of three neural networks
    Details and architectures of three networks. (a) FU-Net; (b) GBU-Net; (c) MBU-Net
    Buildings in satellite remote sensing image and their corresponding labels. (a) Satellite remote sensing image; (b) labeled image
    Part samples of satellite remote sensing data for training
    Visualization result of convolution kernel group parameters of MBU-Net
    Input image for visualizing of MBU-Net interlayer
    Visualization results of 30 feature maps in third layer
    Trends of loss of five networks
    Trends of accuracy of five networks
    Partial input images and output prediction on test set
    • Table 1. Input and output parameters of MBU-Net layer

      View table

      Table 1. Input and output parameters of MBU-Net layer

      Convolution typeInputKernel sizeStridePaddingOutput
      BinaryConv2D256×256×33×3×641Same256×256×64
      BinaryConv2D256×256×643×3×641Same256×256×64
      Pool256×256×642×21128×128×64
      BinaryConv2D128×128×643×3×1281Same128×128×128
      BinaryConv2D128×128×1283×3×1281Same128×128×128
      Pool128×128×1282×2164×64×128
      BinaryConv2D64×64×1283×3×2561Same64×64×256
      BinaryConv2D64×64×2563×3×2561Same64×64×256
      Pool64×64×2562×232×32×256
      BinaryConv2D32×32×2563×3×5121Same32×32×512
      BinaryConv2D32×32×5123×3×5121Same32×32×512
      Pool32×32×5122×216×16×512
      BinaryConv2D256×256×1283×3×641Same256×256×64
      BinaryConv2D256×256×643×3×161Same256×256×16
      Conv2D256×256×161×1×11Same256×256×1
    • Table 2. Results of training loss and accuracy of five networks

      View table

      Table 2. Results of training loss and accuracy of five networks

      AlgorithmTraining lossTraining accuracy /%
      FU-Net0.0199.56
      GBU-Net0.4187.73
      MBU-Net0.1693.94
      Deeplab0.0199.70
      ENet0.2392.21
    • Table 3. Test indexes of five networks on test set

      View table

      Table 3. Test indexes of five networks on test set

      AlgorithmβPA/%P /%R /%F1score /%Memory /MBTime /ms
      FU-Net81.8877.0766.2071.2296.60357
      GBU-Net74.7484.4154.2266.033.0247
      MBU-Net82.3382.7665.5473.153.0249
      Deeplab82.0378.6367.1172.41123.47489
      ENet78.4985.3756.1967.7713.73107
    • Table 4. Comparison of energy consumption of addition and multiplication operations for different types of parameters[32]

      View table

      Table 4. Comparison of energy consumption of addition and multiplication operations for different types of parameters[32]

      OperationEnergy consumption /pJ
      MultiplicationAddition
      8 bit integer0.200.03
      32 bit integer3.100.10
      16 bit floating point1.100.40
      32 bit floating point3.700.90
    Tools

    Get Citation

    Copy Citation Text

    Tianyou Zhu, Feng Dong, Huixing Gong. Remote Sensing Building Detection Based on Binarized Semantic Segmentation[J]. Acta Optica Sinica, 2019, 39(12): 1228002

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: May. 27, 2019

    Accepted: Aug. 13, 2019

    Published Online: Dec. 6, 2019

    The Author Email: Huixing Gong (hxgong@mail.sitp.ac.cn)

    DOI:10.3788/AOS201939.1228002

    Topics