Optics and Precision Engineering, Volume. 31, Issue 4, 517(2023)

Ship detection oriented to compressive sensing measurements of space optical remote sensing scenes

Shuming XIAO1,2, Ye ZHANG1,2、*, Xuling CHANG1,2, and Jianbo SUN1,2
Author Affiliations
  • 1Changchun Institute of Optics, Fine Mechanics and Physics,Chinese Academy of Sciences, Changchun30033, China
  • 2University of Chinese Academy of Sciences, Beijing100039, China
  • show less
    Figures & Tables(25)
    Illustration of the pipeline of the CS-based SORS imaging system to perform ship detection tasks, where digital mirror device (DMD) denotes a measurement matrix in the CS-based imaging system
    Illustration of the overall framework of CS-IM-YOLO, including three parts: CML, IDBN and FPN
    Illustration of compression sampling process
    Illustration of improved Darknet53 backbone network
    Illustration of SENet
    Illustration of FPN
    Illustration of joint optimization process of CS-IM-YOLO
    Part of HRSC2016 dataset
    F1 and PR curves of CS-IM-YOLO for CS measurements with MRs=25%
    Ship detection results or CS measurements in some scenes of the test set
    F1 and PR curves of CS-IM-YOLO for measurements when MRs=25% and 10%
    F1 and PR curves of “Darknet53+FPN” and “IDBN+FPN” for measurements when MRs=25%
    Degradation processing results of scene A in the test set
    Degradation processing results of scene B in the test set
    Reduced resolution processing results of scene A in the test set
    Reduced resolution processing results of scene B in the test set
    • Table 1. Relationship between stride and MRs

      View table
      View in Article

      Table 1. Relationship between stride and MRs

      B×BMRs
      2×225%或10%
    • Table 2. Dataset partitioning

      View table
      View in Article

      Table 2. Dataset partitioning

      数据集场景数量
      训练集1 176
      验证集168
      测试集336
    • Table 3. Experimental environment

      View table
      View in Article

      Table 3. Experimental environment

      系统Ubuntu 18.04
      RAM32.0 GB
      CPU4.10 GHz Intel processor
      GPUGeForce RTX 3070, memory 8 G
      DL框架Pytorch
    • Table 4. Training parameters

      View table
      View in Article

      Table 4. Training parameters

      卷积初始化标准差为0.001的高斯分布28
      OptimizerAdam
      Learn rate10-3
      Batch size8
    • Table 5. Ship detection results on measurements in HRSC2016 dataset

      View table
      View in Article

      Table 5. Ship detection results on measurements in HRSC2016 dataset

      模型PRF1AP
      CS-IM-YOLO91.60%87.59%0.9094.13%
    • Table 6. Ship detection results of CS-IM-YOLO under different MRs

      View table
      View in Article

      Table 6. Ship detection results of CS-IM-YOLO under different MRs

      B×BPRF1AP
      10%90.72%78.47%0.8488.57%
      25%91.60%87.59%0.9094.13%
    • Table 7. Ship detection results on CS measurements of SORS scenes at MRs=25%

      View table
      View in Article

      Table 7. Ship detection results on CS measurements of SORS scenes at MRs=25%

      模型PRF1AP
      Darknet53+FPN89.31%85.40%0.8792.39%
      IDBN+FPN91.60%87.59%0.9094.13%
    • Table 8. Ship detection results on CS measurements of degraded SORS scenes at MRs=25%

      View table
      View in Article

      Table 8. Ship detection results on CS measurements of degraded SORS scenes at MRs=25%

      场景处理PRF1AP
      无退化处理91.60%87.59%0.9094.13 %
      运动模糊90.76%78.83%0.8487.39%
      高斯噪声87.31%85.40%0.8690.51%
      运动模糊+高斯噪声85.54%77.74%0.8186.16%
    • Table 9. Ship detection results on CS measurements of reduced resolution SORS scenes at MRs=25%

      View table
      View in Article

      Table 9. Ship detection results on CS measurements of reduced resolution SORS scenes at MRs=25%

      场景处理PRF1AP
      分辨率/191.60%87.59%0.9094.13 %
      分辨率/489.06%83.21%0.8691.88%
      分辨率/885.29%74.09%0.7981.14%
    Tools

    Get Citation

    Copy Citation Text

    Shuming XIAO, Ye ZHANG, Xuling CHANG, Jianbo SUN. Ship detection oriented to compressive sensing measurements of space optical remote sensing scenes[J]. Optics and Precision Engineering, 2023, 31(4): 517

    Download Citation

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

    Category: Information Sciences

    Received: Mar. 31, 2022

    Accepted: --

    Published Online: Mar. 7, 2023

    The Author Email: Ye ZHANG (yolanda@spirits.ai)

    DOI:10.37188/OPE.20233104.0517

    Topics