• Laser & Optoelectronics Progress
  • Vol. 56, Issue 10, 102802 (2019)
Liang Pei, Yang Liu*, and Lin Gao
Author Affiliations
  • School of Geomatics, Liaoning Technical University, Fuxin, Liaoning 123000, China
  • show less
    DOI: 10.3788/LOP56.102802 Cite this Article Set citation alerts
    Liang Pei, Yang Liu, Lin Gao. Cloud Detectionof ZY-3 Remote Sensing Images Based on Fully Convolutional Neural Network and Conditional Random Field[J]. Laser & Optoelectronics Progress, 2019, 56(10): 102802 Copy Citation Text show less
    Flow chart of proposed algorithm
    Fig. 1. Flow chart of proposed algorithm
    Principle diagram of fully convolutional network
    Fig. 2. Principle diagram of fully convolutional network
    Input images and label pictures. (a) Input image 1; (b) label picture 1; (c) input image 2; (d) label picture 2
    Fig. 3. Input images and label pictures. (a) Input image 1; (b) label picture 1; (c) input image 2; (d) label picture 2
    Comparison of cloud detection results with different methods. (a) Input image; (b) label picture; (c) FCN+CRF; (d) MyFCN; (e) FCN-8s; (f) FCM+SVM
    Fig. 4. Comparison of cloud detection results with different methods. (a) Input image; (b) label picture; (c) FCN+CRF; (d) MyFCN; (e) FCN-8s; (f) FCM+SVM
    From top to bottomFrom bottom to top
    Conv1_1-64n-3×3k-‘relu’
    Conv1_2-64n-3×3k-‘relu’
    Pool1-2×2 max poolingDeconv9-64n-3×3k-‘relu’
    Conv2_1-128n-3×3k-‘relu’Conv8_1-128n-3×3k-‘relu’
    Conv2_2-128n-3×3k-‘relu’Conv8_2-128n-3×3k-‘relu’
    Pool2-2×2 max poolingMerge8 (conv2_2, Deconv8)
    Conv3_1-256n-3×3k-‘relu’Deconv8-128n-3×3k-‘relu’
    Conv3_2-256n-3×3k-‘relu’Conv7_2-256n-3×3k-‘relu’
    Pool3-2×2 max poolingConv7_1-256n-3×3k-‘relu’
    Conv4_1-512n-3×3k-‘relu’Merge7 (conv3_2, Deconv7)
    Conv4_2-512n-3×3k-‘relu’Conv7-256n-3×3k-‘relu’
    Drop4-dropout (0.5)Conv6_2-512n-3×3k-‘relu’
    Pool4-2×2 max poolingConv6_1-512n-3×3k-‘relu’
    Conv5_1-1024n-3×3k-‘relu’Merge6 (drop4, Deconv6)
    Conv5_2-1024n-3×3k-‘relu’Deconv6-512n-3×3k-‘relu’
    Table 1. Structural code mapping in improved fully convolutional neural network algorithm
    ItemFCN+CRFMyFCNFCN-8sFCM+SVM
    Raccuracy /%97.3890.1183.9683.72
    Rprecision /%99.7096.9294.1584.89
    Rrecall /%96.7790.3685.3584.61
    RF1measure /%98.2293.5389.5384.75
    Detection time /s-0.462.24>60
    Table 2. Comparison of accuracy and detection time of cloud area detection by different methods
    Liang Pei, Yang Liu, Lin Gao. Cloud Detectionof ZY-3 Remote Sensing Images Based on Fully Convolutional Neural Network and Conditional Random Field[J]. Laser & Optoelectronics Progress, 2019, 56(10): 102802
    Download Citation