Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Acta Optica Sinica, Volume. 39, Issue 4, 0428004(2019)

Fully Convolutional Network Method of Semantic Segmentation of Class Imbalance Remote Sensing Images

Zhihuan Wu1,2、*, Yongming Gao3, Lei Li4, and Junshi Xue1
Author Affiliations
  • 1 Graduate School, Space Engineering University, Beijing 101416, China
  • 2 63883 Troops, Luoyang, Henan 471000, China
  • 3 School of Space Information, Space Engineering University, Beijing 101416, China
  • 4 Department of Electronic and Optical Engineering, Space Engineering University, Beijing 101416, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (14)
    Equations (0)
    References (0)
    Cited By (8)
    Tools
    Paper Information
    Topics
    Figures & Tables(14)
    Comparison of FCN and CNN based on image blocks. (a) CNN; (b) FCN

    Fig. 1. Comparison of FCN and CNN based on image blocks. (a) CNN; (b) FCN

    Download full size
    Architecture of model system

    Fig. 2. Architecture of model system

    Download full size
    Class distribution of DSTL dataset

    Fig. 3. Class distribution of DSTL dataset

    Download full size
    Labels of image in DSTL dataset

    Fig. 4. Labels of image in DSTL dataset

    Download full size
    Training process. (a) Accuracy; (b) loss function; (c) Jaccard_coef; (d) Jaccard_coef_int

    Fig. 5. Training process. (a) Accuracy; (b) loss function; (c) Jaccard_coef; (d) Jaccard_coef_int

    Download full size
    Results predicted by model

    Fig. 6. Results predicted by model

    Download full size
    Relationship between evaluation value and threshold of each class

    Fig. 7. Relationship between evaluation value and threshold of each class

    Download full size
    Experimental results by proposed method. (a)-(c) Ground truths; (d)-(f) results with adaptive threshold; (g)-(i) results without adaptive threshold

    Fig. 8. Experimental results by proposed method. (a)-(c) Ground truths; (d)-(f) results with adaptive threshold; (g)-(i) results without adaptive threshold

    Download full size
    Experimental results by proposed method. (a)-(c) Results of Patch-based CNN model; (d)-(f) results with adaptive threshold and without data augmentation; (g)-(i) results without adaptive threshold and without data augmentation

    Fig. 9. Experimental results by proposed method. (a)-(c) Results of Patch-based CNN model; (d)-(f) results with adaptive threshold and without data augmentation; (g)-(i) results without adaptive threshold and without data augmentation

    Download full size
    Experimental results of small class. (a)(b) Original images; (c)(d) ground truths; (e)(f) results of proposed method; (g)(h) results of basic U-Net model

    Fig. 10. Experimental results of small class. (a)(b) Original images; (c)(d) ground truths; (e)(f) results of proposed method; (g)(h) results of basic U-Net model

    Download full size
    Comparison of algorithm performance

    Fig. 11. Comparison of algorithm performance

    Download full size

    • Table 1. Specifications of DSTL dataset at different bands

      View table

      Table 1. Specifications of DSTL dataset at different bands

      BandRadiometric resolution /bitSpatial resolution /mSize /(pixel×pixel)
      RGB+P (450-690 nm)110.313348×3392
      M band (400-1040 nm)111.24837×848
      A band (1195-2365 nm)147.50134×136

    • Table 2. Best threshold of each class

      View table

      Table 2. Best threshold of each class

      Classbuildingsmiscroadtracktreescropswaterwaystanding watervehicle largevehicle small
      Threshold0.360.220.510.260.450.390.570.610.310.18

    • Table 3. Comparison of algorithm performance (Jaccard index)

      View table

      Table 3. Comparison of algorithm performance (Jaccard index)

      Algorithmbuildingsmiscroadtracktreescropswaterwaystanding watervehicle largevehicle smallAverage
      Binary logistic0.4840.0150.5000.2010.3930.5390.5750000.271
      Patch-based CNN0.6230.1460.7560.2580.6710.9580.8730.8000.0010.0300.512
      FCN+DA0.6980.1520.8630.4550.7280.9690.9170.8300.3590.1660.614
      FCN+WCE+DA0.7010.1810.8630.4670.7280.9690.9180.8340.3660.1830.621
      FCN+AT+DA0.7020.2280.8630.4650.7280.9690.9180.8320.3620.2260.629
      FCN+WCE+AT0.5990.1380.5680.2190.4510.8780.8290.7070.0420.0470.448
      Proposed method0.7050.2580.8640.4720.7280.9690.9190.8350.3690.2380.636
    Tools

    Get Citation

    Copy Citation Text

    Zhihuan Wu, Yongming Gao, Lei Li, Junshi Xue. Fully Convolutional Network Method of Semantic Segmentation of Class Imbalance Remote Sensing Images[J]. Acta Optica Sinica, 2019, 39(4): 0428004

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: Oct. 22, 2018

    Accepted: Dec. 29, 2018

    Published Online: May. 10, 2019

    The Author Email:

    DOI:10.3788/AOS201939.0428004

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology