Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 57, Issue 20, 201505(2020)

Parallel FPN Algorithm Based on Cascade R-CNN for Object Detection from UAV Aerial Images

Yingjie Liu, Fengbao Yang*, and Peng Hu
Author Affiliations
  • School of Information and Communication Engineering, North University of China, Taiyuan, Shanxi 030051, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (13)
    Equations (0)
    References (21)
    Tools
    Paper Information
    Topics
    Figures & Tables(13)
    Object area versus number of objects

    Fig. 1. Object area versus number of objects

    Download full size
    Number of objects per image versus percentage of number of images

    Fig. 2. Number of objects per image versus percentage of number of images

    Download full size
    Example of actual aerial images

    Fig. 3. Example of actual aerial images

    Download full size
    FPN frame

    Fig. 4. FPN frame

    Download full size
    Parallel FPN frame

    Fig. 5. Parallel FPN frame

    Download full size
    Structure of Cascade R-CNN

    Fig. 6. Structure of Cascade R-CNN

    Download full size
    Change of positive proposals in every stage. (a) IoU threshold is 0.5; (b) IoU threshold is 0.6; (c) IoU threshold is 0.7

    Fig. 7. Change of positive proposals in every stage. (a) IoU threshold is 0.5; (b) IoU threshold is 0.6; (c) IoU threshold is 0.7

    Download full size
    Overall framework of proposed model

    Fig. 8. Overall framework of proposed model

    Download full size
    Visual detection results of aerial images

    Fig. 9. Visual detection results of aerial images

    Download full size

    • Table 1. Comparison of classical algorithms %

      View table

      Table 1. Comparison of classical algorithms %

      AlgorithmBackboneAPAP0.5AP0.75APSAPMAPL
      Retina-Net[19]ResNet-1017.113.27.02.912.417.6
      Faster R-CNN4.610.13.72.77.17.8
      R-FCN[20]7.916.66.54.512.418.1
      FPN w nearest17.037.313.611.425.627.7
      FPN w bilinear17.237.414.011.525.928.0
      Proposed algorithm17.537.514.511.526.228.4
      FPN w nearestResNet-101-v1d20.140.917.714.029.132.5
      FPN w bilinear20.341.018.114.029.332.9
      Proposed algorithm20.641.118.714.129.633.3

    • Table 2. Impact of number of cascading stages on parameters %

      View table

      Table 2. Impact of number of cascading stages on parameters %

      NetworkAPAP0.5AP0.75
      Without Cascade R-CNN20.641.118.7
      With CascadeR-CNNStage 1-225.646.625.4
      Stage 1-326.146.425.7
      Stage 1-425.746.225.3

    • Table 3. Comparison of anchor size of proposed algorithm

      View table

      Table 3. Comparison of anchor size of proposed algorithm

      Anchor schemeAP /%APS /%APM /%APL /%
      A25.017.435.736.8
      B26.118.337.038.3
      C25.818.036.436.4

    • Table 4. Detection result of multi-scale training

      View table

      Table 4. Detection result of multi-scale training

      DatasetMulti-scale trainingAP /%AP0.5 /%AP0.75 /%
      MS COCO×/√26.1/26.746.4/48.125.7/26.1
      VisDrone×/√27.34/27.9852.31/52.5424.84/25.62
    Tools

    Get Citation

    Copy Citation Text

    Yingjie Liu, Fengbao Yang, Peng Hu. Parallel FPN Algorithm Based on Cascade R-CNN for Object Detection from UAV Aerial Images[J]. Laser & Optoelectronics Progress, 2020, 57(20): 201505

    Download Citation

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

    Category: Machine Vision

    Received: Dec. 10, 2019

    Accepted: Feb. 25, 2020

    Published Online: Oct. 17, 2020

    The Author Email: Yang Fengbao (yfengb@163.com)

    DOI:10.3788/LOP57.201505

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology