• Acta Photonica Sinica
  • Vol. 50, Issue 7, 79 (2021)
Dongen GUO1、2, Ying XIA1, Xiaobo LUO1, and Jiangfan FENG1
Author Affiliations
  • 1Chongqing Engineering Research Center for Spatial Big Data Intelligent Technology, Chongqing University of Posts and Telecommunications, Chongqing400065, China
  • 2School of Computer and Software, Nanyang Institute of Technology, Nanyang,Henan473000, China
  • show less
    DOI: 10.3788/gzxb20215007.0710002 Cite this Article
    Dongen GUO, Ying XIA, Xiaobo LUO, Jiangfan FENG. Remote Sensing Image Scene Classification Based on Supervised Contrastive Learning[J]. Acta Photonica Sinica, 2021, 50(7): 79 Copy Citation Text show less
    The structure of the proposed model
    Fig. 1. The structure of the proposed model
    The size of the feature maps on each key node in the model
    Fig. 2. The size of the feature maps on each key node in the model
    The structure of gated self-attention module
    Fig. 3. The structure of gated self-attention module
    Some examples from AID data set
    Fig. 4. Some examples from AID data set
    Some examples from NWPU-RESISC45 data set
    Fig. 5. Some examples from NWPU-RESISC45 data set
    Histogram of classification results on AID dataset with training ratio of 20% and 50%
    Fig. 6. Histogram of classification results on AID dataset with training ratio of 20% and 50%
    Confusion matrix generated under 20% training ratio on the AID dataset
    Fig. 7. Confusion matrix generated under 20% training ratio on the AID dataset
    Confusion matrix generated under 50% training ratio on the AID dataset
    Fig. 8. Confusion matrix generated under 50% training ratio on the AID dataset
    Confusion matrix generated under 20% training ratio on the NWPU-RESISC45 dataset
    Fig. 9. Confusion matrix generated under 20% training ratio on the NWPU-RESISC45 dataset
    DatasetImages per categoryNumber of categoriesTotal imagesSizeTraining ratio
    AID220~4203010 000600×60020%/50%
    NWPU-RESISC457004531 500256×25610%/20%
    Table 1. Data set description
    MethodsYearAID datasetNWPU-RESISC45 dataset
    20% training ratio50% training ratio10% training ratio10% training ratio
    D-CNN2201890.82±0.1696.89±0.1088.07±0.1888.07±0.18
    MSCP18201892.21±0.1796.56±0.1889.03±0.2189.03±0.21
    SF-CNN23201993.60±0.1296.66±0.1190.87±0.1590.87±0.15
    CNN-CapsNet24201993.79±0.1396.63±0.1291.48±0.1991.48±0.19
    MF2Net25201993.82±0.2695.93±0.2391.63±0.1591.63±0.15
    MG-CAP(Bil)26202092.11±0.1595.14±0.1288.48±0.2188.48±0.21
    DDRL-AM27202092.36±0.1096.25±0.0589.22±0.5089.22±0.50
    SCCov28202093.12±0.2596.10±0.1689.42±0.1989.42±0.19
    MG-CAP(Sqr)26202093.34±0.1896.12±0.1289.89±0.1689.89±0.16
    ResNet50EAM29202093.64±0.2596.62±0.1391.08±0.2491.08±0.24
    ResNet101EAM29202094.26±0.1197.06±0.1991.91±0.2291.91±0.22
    Ours--95.66±0.1997.29±0.2292.86±0.2092.86±0.20
    Table 2. Performance comparisons of different methods on AID and NWPU-RESISC45 datasets
    MethodsAID datasetNWPU-RESISC45 dataset
    20% training ratio50% training ratio10% training ratio20% training ratio
    SCRSISC-SC94.52±0.1396.26±0.1191.83±0.1193.78±0.12
    SCRSISC-GSA94.36±0.1796.14±0.1891.38±0.2493.51±0.19
    SCRSISC-Inc_v394.16±0.1596.05±0.2591.16±0.1093.28±0.13
    Ours(SCRSISC)95.66±0.1997.29±0.2292.86±0.2094.73±0.19
    Table 3. Performance comparisons of different variants on two datasets
    MethodsAID datasetNWPU-RESISC45 dataset
    Times/minAccuracyTimes/minAccuracy
    ResNet50EAM296193.64±0.2516893.49±0.17
    ResNet101EAM2910994.26±0.1128294.29±0.09
    Ours (SCRSISC)4595.66±0.1912694.73±0.19
    Table 4. Comparisons of efficiency and accuracy for two datasets at 20% training ratio
    Dongen GUO, Ying XIA, Xiaobo LUO, Jiangfan FENG. Remote Sensing Image Scene Classification Based on Supervised Contrastive Learning[J]. Acta Photonica Sinica, 2021, 50(7): 79
    Download Citation