• Laser & Optoelectronics Progress
  • Vol. 56, Issue 5, 052801 (2019)
Liang Pei1, Yang Liu1、2、*, Hai Tan2, and Lin Gao1
Author Affiliations
  • 1 School of Geomatics, Liaoning Technical University, Fuxin, Liaoning 123000, China
  • 2 Satellite Surveying and Mapping Application Center, NASG, Beijing 100048, China
  • show less
    DOI: 10.3788/LOP56.052801 Cite this Article Set citation alerts
    Liang Pei, Yang Liu, Hai Tan, Lin Gao. Cloud Detection of ZY-3 Satellite Remote Sensing Images Based on Improved Fully Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2019, 56(5): 052801 Copy Citation Text show less
    Schematic of FCNN
    Fig. 1. Schematic of FCNN
    Schematic of principles of FCNN
    Fig. 2. Schematic of principles of FCNN
    Schematic of principle of improved FCNN
    Fig. 3. Schematic of principle of improved FCNN
    Structure diagram of improved FCNN
    Fig. 4. Structure diagram of improved FCNN
    Input images and label pictures. (a) Input image 1; (b) label picture 1; (c) input image 2; (d) label picture 2
    Fig. 5. Input images and label pictures. (a) Input image 1; (b) label picture 1; (c) input image 2; (d) label picture 2
    Convergence curves of Adam algorithm. (a) Loss; (b)accuracy
    Fig. 6. Convergence curves of Adam algorithm. (a) Loss; (b)accuracy
    Convergence curves of SGD algorithm. (a) Loss; (b) accuracy
    Fig. 7. Convergence curves of SGD algorithm. (a) Loss; (b) accuracy
    Comparison of detection results of different methods. (a) Original images; (b) label picture; (c) FCN-2s; (d) FCN-8s;(e) FCN-16s; (f) FCN-32s; (g) FCM+SVM
    Fig. 8. Comparison of detection results of different methods. (a) Original images; (b) label picture; (c) FCN-2s; (d) FCN-8s;(e) FCN-16s; (f) FCN-32s; (g) FCM+SVM
    Types ofnetworkOverallaccuracyMeanaccuracyMeanIU
    FCN-8s0.89880.80100.7396
    FCN-16s0.89650.79630.7339
    FCN-32s0.88910.78670.7185
    Table 1. Accuracy comparison of different network structures
    Types of networkAccuracy /%Precision /%Recall /%F1-measure /%Detection time /s
    FCN-2s90.1196.9290.3693.530.46
    FCN-8s83.9694.1585.3589.532.24
    FCN-16s80.3692.4881.9486.892.30
    FCN-32s76.5288.2979.6783.762.36
    FCM+SVM83.7284.8984.6184.75>60.00
    Table 2. Average accuracy and speed of detection results in different cloud areas
    Liang Pei, Yang Liu, Hai Tan, Lin Gao. Cloud Detection of ZY-3 Satellite Remote Sensing Images Based on Improved Fully Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2019, 56(5): 052801
    Download Citation