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. 60, Issue 2, 0228003(2023)

Multiscale Object Detection Algorithm for Satellite Remote-Sensing Images

Jianhong Xiang1,2, Zhenxing Chen1,2, and Linyu Wang1,2、*
Author Affiliations
  • 1College of Information & Communication Engineering, Harbin Engineering University, Harbin 150001, Heilongjiang, China
  • 2Key Laboratory of Advanced Ship Communication and Information Technology, Harbin Engineering University, Harbin 150001, Heilongjiang, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (15)
    Equations (7)
    References (25)
    Cited By (1)
    Tools
    Paper Information
    Topics
    Figures & Tables(15)
    YOLOv5 network structure diagram

    Fig. 1. YOLOv5 network structure diagram

    Download full size
    Mosaic data augmentation. (a) (b) (c) Effect after use

    Fig. 2. Mosaic data augmentation. (a) (b) (c) Effect after use

    Download full size
    CAM structure diagram

    Fig. 3. CAM structure diagram

    Download full size
    SAM structure diagram

    Fig. 4. SAM structure diagram

    Download full size
    Structure comparison between FPN+PAN and WMFPN

    Fig. 5. Structure comparison between FPN+PAN and WMFPN

    Download full size
    Fusion network structure of P4 layer

    Fig. 6. Fusion network structure of P4 layer

    Download full size
    Output feature map comparison between proposed network and original network. (a) Original picture; (b) original network feature map; (c) proposed network feature map

    Fig. 7. Output feature map comparison between proposed network and original network. (a) Original picture; (b) original network feature map; (c) proposed network feature map

    Download full size
    YOLOv5-WMFPN structure diagram

    Fig. 8. YOLOv5-WMFPN structure diagram

    Download full size
    DIOR optical remote sensing image dataset categories. (a) Airplane; (b) airport; (c) baseball field; (d) basketball count; (e) bridge; (f) chimney; (g) dam; (h) expressway service area; (i) expressway toll station; (j) golf course; (k) ground track field; (l) harbor; (m) overpass; (n) ship; (o) stadium; (p) storage tank; (q) tennis court; (r) train station; (s) vehicle; (t) wind mill

    Fig. 9. DIOR optical remote sensing image dataset categories. (a) Airplane; (b) airport; (c) baseball field; (d) basketball count; (e) bridge; (f) chimney; (g) dam; (h) expressway service area; (i) expressway toll station; (j) golf course; (k) ground track field; (l) harbor; (m) overpass; (n) ship; (o) stadium; (p) storage tank; (q) tennis court; (r) train station; (s) vehicle; (t) wind mill

    Download full size
    Airplane and wind mill object detection. (a) (b) YOLOv5m airplane; (c) (d) YOLOv5m wind mill; (e) (f) YOLOv5m-WMFPN airplane; (g) (h) YOLOv5m-WMFPN wind mill

    Fig. 10. Airplane and wind mill object detection. (a) (b) YOLOv5m airplane; (c) (d) YOLOv5m wind mill; (e) (f) YOLOv5m-WMFPN airplane; (g) (h) YOLOv5m-WMFPN wind mill

    Download full size
    Vehicle and ship object detection. (a) (b) YOLOv5m vehicle; (c) (d) YOLOv5m ship; (e) (f) YOLOv5m-WMFPN vehicle; (g) (h) YOLOv5m-WMFPN ship

    Fig. 11. Vehicle and ship object detection. (a) (b) YOLOv5m vehicle; (c) (d) YOLOv5m ship; (e) (f) YOLOv5m-WMFPN vehicle; (g) (h) YOLOv5m-WMFPN ship

    Download full size

    • Table 1. Comparison of various APs between YOLOv5-WMFPN and YOLOv5m algorithm

      View table

      Table 1. Comparison of various APs between YOLOv5-WMFPN and YOLOv5m algorithm

      Algorithmc1c2c3c4c5c6c7c8c9c10
      YOLOv5m78.478.273.888.14577.564.36259.677.9
      YOLOv5m-WMFPN89.18374.491.9478264.663.66579.2
      Algorithmc11c12c13c14c15c16c17c18c19c20
      YOLOv5m72.660.458.488.965.676.686.76355.177
      YOLOv5m-WMFPN77.363.161.189.269.377.489.764.858.681.3

    • Table 2. Overall effect comparison between YOLOv5m-WMFPN and YOLOv5m

      View table

      Table 2. Overall effect comparison between YOLOv5m-WMFPN and YOLOv5m

      AlgorithmP /%R /%mAP@0.5 /%mAP@［.5：.95］ /%Detection speed /（frame·s-1）
      YOLOv5m86.165.370.546.6113.6
      YOLOv5m-WMFPN84.968.273.650.5102.0

    • Table 3. Ablation experiment results

      View table

      Table 3. Ablation experiment results

      ExperimentMFPNWMFPNCBAMmAP /%Detection speed /（frame·s-1）
      170.5113.6
      2√71.2109.4
      3√71.5107.3
      4√√73.3106.1
      5√√√73.6102.0

    • Table 4. Performance comparison of each algorithm

      View table

      Table 4. Performance comparison of each algorithm

      AlgorithmBackboneNeckmAP /%
      Faster RCNNVGG1657.6
      Faster RCNN-FPNResNet50FPN67.4
      SSDVGG1654.3
      Yolov3Darknet53FPN64.2
      Yolov3-SPPDarknet53FPN67.5
      Efficientdet-d2EfficientnetBiFPN55..5
      Efficientdet-d4EfficientnetBiFPN63.9
      YOLOv5sC3CSPFPN+PAN67.2
      YOLOv5mC3CSPFPN+PAN70.5
      YOLOv5m-WMFPNC3CSP-CBAMWMFPN73.6
    Tools

    Get Citation

    Copy Citation Text

    Jianhong Xiang, Zhenxing Chen, Linyu Wang. Multiscale Object Detection Algorithm for Satellite Remote-Sensing Images[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0228003

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: Oct. 8, 2021

    Accepted: Nov. 16, 2021

    Published Online: Jan. 6, 2023

    The Author Email: Linyu Wang (wanglinyu@hrbeu.edu.cn)

    DOI:10.3788/LOP212670

    Topics

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