Laser & Optoelectronics Progress, Volume. 58, Issue 16, 1601002(2021)

Objects Detection from High-Resolution Remote Sensing Imagery Using Training-Optimized YOLOv3 Network

Yun Yang... Longwei Li*, Siyan Gao, Han Bai and Wancheng Jiang |Show fewer author(s)
Author Affiliations
  • School of Geological Engineering and Surveying and Mapping, Chang’an University, Xi’an, Shaanxi 710054, China
  • show less
    Figures & Tables(9)
    Principle of YOLO model for object detection
    Our optimized YOLOv3 model training flowchart
    Three objects contained in the RSOD dataset
    RSOD dataset augmentation processing. (a)Image flipping along X axis; (b)image cropping; (c) image rotation; (d) image saturation adjustment
    Labeling sample of dataset using Labeling software. (a) Original image; (b) labeled image (true object bounding box)
    Comparison of the three basic sizes of anchor boxes before and after our training optimization. (a) Anchor boxes from traditional YOLOv3 model; (b) anchor boxes from our clustered augmented dataset
    Comparison of the detection results between the traditional and our optimized YOLOv3 models. (a)--(c) Traditional YOLOv3 model; (d)--(f) our optimized YOLOv3 model
    • Table 1. YOLOv3 model object detection result before and after optimization

      View table

      Table 1. YOLOv3 model object detection result before and after optimization

      SceneDetection number of traditional modelDetection number of optimized model
      AircraftOverpassPlaygroundAircraftOverpassPlayground
      Aircraft3507539076
      Overpass161704181723
      Playground14101601511162
    • Table 2. Evaluation results using the traditional and optimized YOLOv3 models

      View table

      Table 2. Evaluation results using the traditional and optimized YOLOv3 models

      ParameterTraditional YOLOv3 modelOur optimized YOLOv3 model
      AircraftOverpassPlaygroundAircraftOverpassPlayground
      Precision0.9210.9090.9470.9240.9070.947
      Recall0.7370.7390.7620.8210.7480.771
      PA,c0.8190.8150.8440.8690.8200.850
      mAP0.8260.846
    Tools

    Get Citation

    Copy Citation Text

    Yun Yang, Longwei Li, Siyan Gao, Han Bai, Wancheng Jiang. Objects Detection from High-Resolution Remote Sensing Imagery Using Training-Optimized YOLOv3 Network[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1601002

    Download Citation

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

    Category: Atmospheric Optics and Oceanic Optics

    Received: Sep. 25, 2020

    Accepted: Dec. 14, 2020

    Published Online: Aug. 19, 2021

    The Author Email: Li Longwei (1049730716@qq.com)

    DOI:10.3788/LOP202158.1601002

    Topics