Optics and Precision Engineering, Volume. 29, Issue 11, 2672(2021)

Recognition of small targets in remote sensing image using multi-scale feature fusion-based shot multi-box detector

Xin CHEN1...2, Min-jie WAN1,2,*, Chao MA1,2, Qian CHEN1,2, and Guo-hua GU12,* |Show fewer author(s)
Author Affiliations
  • 1School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing20094, China
  • 2Jiangsu Key Laboratory of Spectral Imaging & Intelligent Sense, Nanjing University of Science and Technology, Nanjing10094, China
  • show less
    Figures & Tables(15)
    SSD network framework
    Structure of improved SSD network
    Process of feature fusion enhancement
    Schematic diagram of FPN bonding
    Schematic diagram of channel attention mechanism
    Schematic diagram of priori frame generation for SSD
    Detection results of different algorithms
    Detection results of SSD algorithm
    Detection results of proposed improved SSD algorithm
    Detection results for different attitudes and types of aircraft targets in air
    • Table 1. Dimensions of original and improved prior frames

      View table
      View in Article

      Table 1. Dimensions of original and improved prior frames

      特征图原本先验框改进后先验框
      Conv4_3_13018
      Fc7_16036
      Conv8_2_111193
      Conv9_2162150
      Conv10_2213207
      Conv11_2264264
    • Table 2. Definition of boundary area and number of targets based on instance size distribution

      View table
      View in Article

      Table 2. Definition of boundary area and number of targets based on instance size distribution

      目标尺度数量
      小目标(0,602]1 236
      中目标(602,1202]1 864
      大目标(1202,+∞]221
    • Table 3. Experimental results of different algorithms on remote sensing aircraft dataset

      View table
      View in Article

      Table 3. Experimental results of different algorithms on remote sensing aircraft dataset

      检测方法基础网络

      AP/%

      (IOU=0.5)

      帧频/(frame·s-1
      Faster R-CNNVGG1687.45.2
      Faster R-CNNResNet-10188.63.6
      YOLOGoogleNet80.533.5
      SSDVGG1685.930.2
      DSSDResNet-10186.212
      FSSDVGG1687.521.2
      Proposed methodVGG1690.224.3
    • Table 4. Results of ablation experiments for each module

      View table
      View in Article

      Table 4. Results of ablation experiments for each module

      SSD特片融合机制通道注意力模块

      先验框

      改进

      AP/%

      (IOU=0.5)

      帧频/(frame·s-1
      85.930.2
      88.725.3
      87.228.5
      86.530.1
      89.824.8
      90.224.3
    • Table 5. Ablation experimental results of feature fusion module

      View table
      View in Article

      Table 5. Ablation experimental results of feature fusion module

      Conv4_3_1Fc7_1Conv8_2_1

      AP/%

      (IOU=0.5)

      Conv3_3Conv4_3Conv9_1Conv4_2Fc7Conv10_1Fc6Conv8_2Conv11_1
      85.9
      87.3
      88.1
      88.7
    Tools

    Get Citation

    Copy Citation Text

    Xin CHEN, Min-jie WAN, Chao MA, Qian CHEN, Guo-hua GU. Recognition of small targets in remote sensing image using multi-scale feature fusion-based shot multi-box detector[J]. Optics and Precision Engineering, 2021, 29(11): 2672

    Download Citation

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

    Category: Information Sciences

    Received: Mar. 27, 2021

    Accepted: --

    Published Online: Dec. 10, 2021

    The Author Email: WAN Min-jie (minjiewan1992@njust.edu.cn), GU Guo-hua (gghnjust@mail.njust.edu.cn)

    DOI:10.37188/OPE.20212911.2672

    Topics