Laser & Optoelectronics Progress, Volume. 62, Issue 6, 0628002(2025)

Remote-Sensing Image Detection Method Based on Contextual Awareness and Sparse Feature Fusion

Quan Feng*, Liang Luo, and Xiaoqian Zhang
Author Affiliations
  • College of Information Engineering, Southwest University of Science and Technology, Mianyang 621010, Sichuan , China
  • show less
    Figures & Tables(9)
    Overall network structure
    Context aware unit and MSDA Transformer. (a) Context aware unit; (b) MSDA Transformer
    Sparse feature fusion strategy
    Structures of GSConv, GSBottleneck and VoGSCSPC. (a) GSConv; (b) GSBottleneck; (c) VoGSCSPC
    Detection results on NWPU VHR-10 and DIOR datasets
    • Table 1. Experimental parameters

      View table

      Table 1. Experimental parameters

      Experimental parameterParameter quantity
      Epoch200
      Batch size16
      Input size640×640
      OptimizerSGD
      Momentum0.937
      Learning rate0.01
      Weight decay0.0005
    • Table 2. Ablation results on NWPU VHR-10 dataset

      View table

      Table 2. Ablation results on NWPU VHR-10 dataset

      No.CAUCFFSSlim-NeckmAP50 /%APS /%APM /%APL /%FLOPs /109Params /106
      190.233.950.451.728.711.1
      291.135.353.254.631.111.3
      391.635.753.955.431.111.3
      491.836.153.055.827.810.4
    • Table 3. Comparative experimental results on the NWPU VHR-10 dataset

      View table

      Table 3. Comparative experimental results on the NWPU VHR-10 dataset

      MethodmAP50C1C2C3C4C5C6C7C8C9C10
      SSD80.384.671.585.388.489.371.299.486.464.962.6
      Faster R-CNN82.498.679.990.689.587.366.395.480.667.069.1
      Mask R-CNN83.985.282.388.190.090.175.297.689.667.673.6
      TOOD86.8100.089.789.099.691.472.6100.086.257.182.8
      YOLOv5s88.499.990.297.299.974.467.799.7100.069.386.2
      YOLOv8s90.297.592.598.397.994.174.9100.095.965.286.2
      Proposed method91.899.691.798.797.791.685.799.889.373.390.2
    • Table 4. Comparative experimental results on the DIOR dataset

      View table

      Table 4. Comparative experimental results on the DIOR dataset

      MethodmAP50C1C2C3C4C5C6C7C8C9C10
      C11C12C13C14C15C16C17C18C19C20
      SSD58.459.572.772.475.748.165.856.663.353.065.3
      68.649.428.259.261.246.676.355.127.464.7
      Faster R-CNN54.253.653.278.866.250.170.962.369.055.268.1
      56.950.228.327.772.039.875.238.623.645.4
      Mask R-CNN60.653.872.663.278.450.266.255.966.155.370.6
      69.244.330.668.055.350.681.156.245.678.3
      TOOD71.690.173.391.981.438.789.254.966.258.875.1
      75.254.461.675.886.180.787.550.868.572.9
      YOLOv5s74.492.973.793.283.148.788.656.971.370.369.5
      79.466.963.994.491.681.390.955.275.880.3
      YOLOv8s76.993.573.194.783.548.089.664.272.270.575.8
      76.169.865.494.096.182.392.550.574.382.8
      Proposed method79.593.681.294.282.350.990.967.774.468.480.2
      82.872.667.394.094.582.692.857.576.082.9
    Tools

    Get Citation

    Copy Citation Text

    Quan Feng, Liang Luo, Xiaoqian Zhang. Remote-Sensing Image Detection Method Based on Contextual Awareness and Sparse Feature Fusion[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0628002

    Download Citation

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

    Category: Remote Sensing and Sensors

    Received: Jul. 25, 2024

    Accepted: Sep. 3, 2024

    Published Online: Mar. 13, 2025

    The Author Email:

    DOI:10.3788/LOP241744

    CSTR:32186.14.LOP241744

    Topics