Acta Optica Sinica, Volume. 44, Issue 12, 1228002(2024)

Synthetic Aperture Radar Ship Detection Method Based on Highly Efficient Aggregated Feature Enhancement Network

Huilin Shan1,2, Wenxing Liu1, Xingtao Wang1, Xiangwei Fu1, Changshuai Li2, and Yinsheng Zhang1,2、*
Author Affiliations
  • 1School of Electronics & Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu , China
  • 2School of Electronic & Information Engineering, Wuxi University, Wuxi 214105, Jiangsu , China
  • show less
    Figures & Tables(15)
    EAFENet structure
    EAFENet FPN model
    Schematic diagram of CBAM
    Schematic diagram of EL-CB
    Inception NeXt module
    GE-FPN model
    Comparison of real detection effects between YOLOv7 and EAFENet. (a) Real values; (b) YOLOv7 detection results; (c) EAFENet detection results
    Comparison of detection effects by different algorithms. (a) SSD; (b) Faster R-CNN; (c) CenterNet; (d) EAFENet
    Comparison of loss for EAFENet and YOLOv7
    Comparison of mAP0.5 for EAFENet and YOLOv7
    Comparison of mAP0.50∶0.95 for EAFENet and YOLOv7
    • Table 1. Parameter configurations of hardware and software platform

      View table

      Table 1. Parameter configurations of hardware and software platform

      ConfigurationModel
      CPU (central processing unit)Intel(R) Xeon(R) CPU E5-2696 v3 @ 2.30 GHz
      GPU (graphic processing unit)NVIDIA GeForce RTX 3060
      Operating systemWindows 10
      Running memory64 Gb
      Video memory12 Gb
      Language and frameworkAnconda3 + Python3.8 + PyTorch1.13.0
    • Table 2. Training parameter configurations

      View table

      Table 2. Training parameter configurations

      Training parameterConfiguration
      Momentum parameter0.937
      Initial learning rate0.001
      Batch size8
      Epoch300
      OptimizerSGD (stochastic gradient descent optimizer)
    • Table 3. Comparison of the effects of various improvement points in ablation experiment

      View table

      Table 3. Comparison of the effects of various improvement points in ablation experiment

      ModelP /%R /%mAP0.5 /%mAP0.50∶0.95 /%
      YOLOv792.988.592.555.7
      YOLOv7+CBAM92.293.296.461.9
      YOLOv7+EL-CB93.693.497.262.3
      YOLOv7+Inception NeXt93.693.396.561.2
      YOLOv7+EL-CB+Inception NeXt94.496.197.965.0
      YOLOv7+EL-CB+InceptionNeXt+GE-FPN95.497.498.965.6
    • Table 4. Comparison results of improved model and mainstream models

      View table

      Table 4. Comparison results of improved model and mainstream models

      ModelP /%R /%mAP0.5 /%FPS /(frame/s)
      CenterNet87.2078.6290.3423.15
      SSD84.1089.3486.0146.13
      Faster R-CNN82.7088.6289.1724.47
      YOLOv593.3289.9694.2533.88
      YOLOv792.9088.5092.5036.18
      Ours95.4094.7098.9031.40
    Tools

    Get Citation

    Copy Citation Text

    Huilin Shan, Wenxing Liu, Xingtao Wang, Xiangwei Fu, Changshuai Li, Yinsheng Zhang. Synthetic Aperture Radar Ship Detection Method Based on Highly Efficient Aggregated Feature Enhancement Network[J]. Acta Optica Sinica, 2024, 44(12): 1228002

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: Jul. 18, 2023

    Accepted: Oct. 10, 2023

    Published Online: May. 23, 2024

    The Author Email: Yinsheng Zhang (yorkzhang@nuist.edu.cn)

    DOI:10.3788/AOS231285

    CSTR:32393.14.AOS231285

    Topics