Laser & Optoelectronics Progress, Volume. 57, Issue 10, 101017(2020)

Remote Sensing Image Retrieval Based on Regional Attention Mechanism

Yanfei Peng**, Jinye Mei*, Kaixin Wang, Lingling Zi, and Yu Sang
Author Affiliations
  • School of Electronic and Information Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
  • show less
    Figures & Tables(14)
    Framework of the retrieval system
    Regional scale of R-MAC
    Framework of ResNet
    Framework diagram of our network
    UCM dataset
    SIRI dataset
    Retrieval effect when return 20 images
    Retrieval effect when return 40 images
    Comparison of average recalls. (a) UCM dataset; (b) SIRI dataset
    Average precision at different scales
    • Table 1. Comparison of mAP of different methodsunit: %

      View table

      Table 1. Comparison of mAP of different methodsunit: %

      MethodVGG16ResNet101DenseNet121Ourmethod
      UCM71.275.978.996.8
      SIRI70.075.774.488.6
      Average70.675.876.792.7
    • Table 2. Comparison of mAP of different image sizes

      View table

      Table 2. Comparison of mAP of different image sizes

      Image size1.00.90.80.70.60.5
      mAP /%88.685.384.782.177.370.9
    • Table 3. Average accuracy of different methods on the UCM datasetunit: %

      View table

      Table 3. Average accuracy of different methods on the UCM datasetunit: %

      MethodVGG16DenseNet121ResNet101Our method
      RANRAN+MDRAN+MD+QE
      Agricultural928681979797
      Airplane698384959695
      Baseball diamond433846100100100
      Beach8185909899100
      Building234446697365
      Chaparral929995100100100
      Dense residential294627778582
      Forest809284100100100
      Freeway434430888992
      Golf course334151949597
      Harbor4375759898100
      Intersection264540969598
      Medium residential377855808280
      Mobile home park698039949595
      Overpass486443949592
      Parking lot627760999395
      River254250888593
      Runway576549878396
      Sparse residential494541798585
      Storage tanks253125928695
      Tennis court3836339899100
      Average50.761.754.591.691.993.2
    • Table 4. Average accuracy of different methods on the SIRI datasetunit: %

      View table

      Table 4. Average accuracy of different methods on the SIRI datasetunit: %

      MethodVGG16DenseNet121ResNet101Our method
      RANRAN+MDRAN+MD+QE
      Agriculture597329979599
      Commercial515967878994
      Harbor505465606267
      Idle land274339858588
      Industrial424966787375
      Meadow293227646969
      Overpass718189979798
      Park354344777983
      Pond394946666276
      Residential516048878790
      River343035899095
      Water939596979898
      Average48.455.754.382.082.286.0
    Tools

    Get Citation

    Copy Citation Text

    Yanfei Peng, Jinye Mei, Kaixin Wang, Lingling Zi, Yu Sang. Remote Sensing Image Retrieval Based on Regional Attention Mechanism[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101017

    Download Citation

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

    Category: Image Processing

    Received: Sep. 12, 2019

    Accepted: Oct. 22, 2019

    Published Online: May. 8, 2020

    The Author Email: Yanfei Peng (pengyf75@126.com), Jinye Mei (1113417696@qq.com)

    DOI:10.3788/LOP57.101017

    Topics