Acta Optica Sinica, Volume. 38, Issue 6, 0615004(2018)

Aerial Target Detection Based on Improved Faster R-CNN

Xiaoyu Feng*, Wei Mei, and Dashuai Hu
Author Affiliations
  • Department of Electronics and Optics Engineering, Army Engineering University, Shijiazhuang, Hebei 050003, China
  • show less
    Figures & Tables(13)
    Frame of Faster R-CNN
    Accumulation of dilation
    Pooling of feature map
    Regional amplification
    Influences of scale and occlusion on confidences
    False positive with high confidence
    Frame of improved Faster R-CNN
    Examples of detection results
    Examples of detection results of M4
    • Table 1. Dataset of aerial targets

      View table

      Table 1. Dataset of aerial targets

      ContentFighterHelicopterAerobusBirdOthers
      Video11 sequences(about 7000 frames)7 sequences(about 6000 frames)10 sequences(about 9000 frames)4 sequences(about 1000 frames)6 sequences(about 4000 frames)
      Image300300300300300
    • Table 2. Result on test set1 %

      View table

      Table 2. Result on test set1 %

      ModelFighterHelicopterAerobusBirdOthersmAP
      M090.7691.0492.3563.7172.7682.12
      M293.4296.6794.2275.8195.7691.18
      M391.5392.6493.1969.3778.0784.96
      M498.7299.3797.4298.6199.7698.78
    • Table 3. Result on test set2%

      View table

      Table 3. Result on test set2%

      ModelFighterHelicopterAerobusBirdOthersmAP
      M0000000
      M154.0568.6662.2639.6185.3661.99
    • Table 4. Detection performances on class aero of VOC

      View table

      Table 4. Detection performances on class aero of VOC

      DetectorFaster R-CNN[9]SSD299[10]SSD443[10]YOLO[11]YOLOv2 544[11]M4
      AP /%84.971.278.677.086.390.7%
      Speed /(frame·s-1)8.34619454025
    Tools

    Get Citation

    Copy Citation Text

    Xiaoyu Feng, Wei Mei, Dashuai Hu. Aerial Target Detection Based on Improved Faster R-CNN[J]. Acta Optica Sinica, 2018, 38(6): 0615004

    Download Citation

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

    Category: Machine Vision

    Received: Dec. 6, 2017

    Accepted: --

    Published Online: Jul. 9, 2018

    The Author Email: Feng Xiaoyu (826782445@qq.com)

    DOI:10.3788/AOS201838.0615004

    Topics