Acta Photonica Sinica, Volume. 50, Issue 11, 1128001(2021)

Remote Sensing Images Target Detection Based on Adjustable Parameter and Receptive field

Nan LIU1... Zhaoyong MAO2, Yichen WANG2 and Junge SHEN2,* |Show fewer author(s)
Author Affiliations
  • 1School of Marine Science and Technology,Northwestern Polytechnical University,Xi'an 710072,China
  • 2Unmanned System Research Institute,Northwestern Polytechnical University,Xi'an 710072,China
  • show less
    Figures & Tables(13)
    Faster R-CNN overall framework
    ASDN overall framework
    Average IOU - cluster number curve
    Model mAP and FPS vs. P conditions
    AP values for the DIOR dataset
    Visual detection results of ASDN512.For each category,ASDN512 is shown on the left and ASDN32 on the right
    PR curves
    • Table 1. Number of instances of each category in DIOR

      View table
      View in Article

      Table 1. Number of instances of each category in DIOR

      CategoryNumberCategoryNumber
      Airplane10 104Ground Track Field3 038
      Airport1 327Harbor5 509
      Baseball Field5 817Overpass3 114
      Basketball Court3 225Ship62 400
      Bridge3 967Stadium1 268
      Chimney1 681Storage Tank26 414
      Dam1 049Tennis Court12 266
      Expressway Service Area2 165Train Station1 011
      Expressway Toll Station1 298Vehicle40 370
      Golf Field1 086Wind Mill5 363
    • Table 2. Confusion matrix

      View table
      View in Article

      Table 2. Confusion matrix

      ActualPredict
      PositiveNegative
      PositiveTrue Positive(TP)False Negative(FN)
      NegativeFalse Positive(FP)True Negative(TN)
    • Table 3. Comparison of the accuracy and receptive field of the combination of different dilation ratios

      View table
      View in Article

      Table 3. Comparison of the accuracy and receptive field of the combination of different dilation ratios

      Dilation ratios(1,2,3)(1,2,5)(1,3,5)(3,5,7)
      mAP/%67.567.868.167.4
      Receptive field188220236332
    • Table 4. Object classes in the DIOR dataset

      View table
      View in Article

      Table 4. Object classes in the DIOR dataset

      C1C2C3C4C5C6C7C8C9C10
      AirplaneAirportBaseball fieldBasketball courtBridgeChimneyDam

      Expressway

      service area

      Expressway

      toll station

      Golf field
      C11C12C13C14C15C16C17C18C19C20
      Ground track fieldHarborOverpassShipStadiumStorage tankTennis courtTrain stationVehicleWind mill
    • Table 5. Comparison of detection accuracy of different algorithms on DIOR

      View table
      View in Article

      Table 5. Comparison of detection accuracy of different algorithms on DIOR

      Model*RICNN21FaterRCNN11*Fater⁃RCNN+FPN22CBD⁃E23

      ASDN512

      (Ours)

      *YOLO⁃V37*SSD8

      *CornerNet

      24

      *Retinanet

      25

      ASDN32

      (Ours)

      BackboneVGG16VGG16

      ResNet⁃

      101

      ResNet⁃

      101

      VGG16

      Darknet⁃

      53

      VGG16Hourglass⁃104

      ResNet‐

      101

      VGG16
      C139.161.55454.265.972.259.558.860.263.9
      C26172.574.57773.529.272.784.27273.8
      C360.168.363.371.573.77472.47270.671.8
      C466.381.980.787.181.778.675.780.880.581
      C525.345.144.844.64831.229.746.443.646.3
      C663.373.172.575.473.569.765.875.372.373.4
      C741.157.66063.556.626.956.664.361.456.3
      C851.770.875.676.275.148.663.581.672.173.4
      C936.663.962.365.36654.453.176.366.766.2
      C1055.974.67679.375.931.165.379.57274.7
      C1158.979.876.879.577.261.168.679.573.475.2
      C1243.543.546.447.551.744.949.426.145.351.1
      C13395957.259.35949.748.160.656.958.4
      C149.147.671.869.176.787.459.237.671.776.2
      C1561.167.968.369.768.270.66170.770.467.4
      C1619.142.453.864.363.868.746.645.26260.2
      C1763.581.181.184.581.787.376.38480.981.4
      C1846.153.559.559.46129.455.157.15758.7
      C1911.43443.144.745.848.327.44347.245.8
      C2031.583.381.283.186.378.765.775.984.583.1
      mAP/%44.263.165.167.868.157.158.664.966.166.9
    • Table 6. Ablation experiments

      View table
      View in Article

      Table 6. Ablation experiments

      ModelKDPRFACNAAFCmAP%FPSParameters
      Faster RCNN×××××59.519.6136.85M
      FRCNN32××××58.834.440.76M
      FRCNN512××××59.419.2139.31M
      FRCNN512C×××59.522.275.13M
      FRCNN K××××64.919.6136.85M
      FRCNN K-×××67.518.8136.85M
      FRCNNK -+××68.118.1136.85M
      ASDN3266.932.321.44M
      ASDN51268.120.475.13M
    Tools

    Get Citation

    Copy Citation Text

    Nan LIU, Zhaoyong MAO, Yichen WANG, Junge SHEN. Remote Sensing Images Target Detection Based on Adjustable Parameter and Receptive field[J]. Acta Photonica Sinica, 2021, 50(11): 1128001

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: May. 7, 2021

    Accepted: Jul. 6, 2021

    Published Online: Dec. 2, 2021

    The Author Email: SHEN Junge (shenjunge@nwpu.edu.cn)

    DOI:10.3788/gzxb20215011.1128001

    Topics