Optics and Precision Engineering, Volume. 31, Issue 18, 2723(2023)

Joint self-attention and branch sampling for object detection on drone imagery

Yunzuo ZHANG1...2,*, Cunyu WU1, Yameng LIU1, Tian ZHANG1 and Yuxin ZHENG1 |Show fewer author(s)
Author Affiliations
  • 1School of Information Science and Technology, Shijiazhuang Tiedao University, Shijiazhuang050043, China
  • 2Hebei Key Laboratory of Electromagnetic Environmental Effects and Information Processing, Shijiazhuang Tiedao University, Shijiazhuang050043, China
  • show less
    Figures & Tables(12)
    Overall structure of the algorithm proposed in this paper
    Structure diagram of NRCS
    Structure diagram of DBS-FU
    Structure diagram of FE-IR
    Comparison of detection results of different algorithms on VisDrone2019
    Comparison of detection results of different algorithms on UAVDT
    • Table 1. Comparison of the results of the baseline and the algorithm in this paper on VisDrone2019

      View table
      View in Article

      Table 1. Comparison of the results of the baseline and the algorithm in this paper on VisDrone2019

      Method指标PedestrianPeopleBicycleCarVanTruckTricycleAwning-tricycleBusMotorALL
      Baseline-smAP5066.853.735.888.456.345.242.421.564.664.353.9
      mAP33.922.516.657.739.930.624.313.646.831.631.7
      Ours-smAP5072.158.042.790.561.053.848.625.272.468.859.3
      mAP38.426.121.667.645.036.029.216.054.636.437.1
      Baseline-mmAP5069.456.039.789.459.152.646.826.371.466.057.7
      mAP34.224.519.767.544.234.328.416.251.933.835.4
      Ours-mmAP5073.660.345.491.162.055.250.629.779.470.361.8
      mAP39.427.123.269.146.539.130.418.359.236.938.9
    • Table 2. Comparison of results of different algorithm models on VisDrone2019

      View table
      View in Article

      Table 2. Comparison of results of different algorithm models on VisDrone2019

      MethodmAP50mAPmAP75Ap-smallAP-midAP-large
      ClusDet1056.232.431.6---
      DSHNet1151.830.330.9---
      HRDNet1262.035.535.1---
      Yolov5-s53.731.731.422.043.749.5
      Yolov5-m58.635.436.827.647.652.4
      Yolov72257.335.637.126.546.050.3
      YoloX53.531.431.722.541.548.5
      MobileNetv355.432.932.924.544.349.5
      MobileViT55.533.333.724.944.241.8
      mSODANet2755.936.937.4---
      Ours-s59.337.138.329.248.551.6
      Ours-m62.138.939.531.251.352.5
    • Table 3. Comparison of the results of the baseline and the algorithm in this paper on UAVDT

      View table
      View in Article

      Table 3. Comparison of the results of the baseline and the algorithm in this paper on UAVDT

      Method指标CarTruckBusAll
      Baseline-smAP5073.015.226.638.3
      mAP41.28.115.921.7
      Ours-smAP5073.319.439.744.1
      mAP40.910.923.024.9
      Baseline-mmAP5072.18.5540.540.4
      mAP39.64.8622.922.5
      Ours-mmAP5073.718.040.944.2
      mAP42.311.224.125.8
    • Table 4. Comparison of results of different algorithm models on UAVDT

      View table
      View in Article

      Table 4. Comparison of results of different algorithm models on UAVDT

      MethodmAP50mAPmAP75
      ClusDet1026.513.912.5
      DMNet2824.614.716.3
      GLSAN2930.519.021.7
      CDMNet3029.116.818.5
      DSHNet1130.417.819.7
      Yolov5-s38.321.722.6
      Yolov5-m41.422.624.1
      PRDet3134.119.821.3
      UFPMP-Det[32]38.724.628.0
      Ours-s44.124.927.3
      Ours-m44.725.328.1
    • Table 5. Performance comparison of different sampling methods

      View table
      View in Article

      Table 5. Performance comparison of different sampling methods

      MethodUp-samplingDown-sampling
      NearestBilinearDBUSConvPoolingDBDS
      mAP5053.754.154.553.753.554.8
      mAP31.732.132.831.732.033.2
    • Table 6. Ablation experiment result

      View table
      View in Article

      Table 6. Ablation experiment result

      BaselineFourheadOur-FourheadNRCSDBUSDBDSFE-IR(3)ParmGFLOPsmAP50(%)mAP(%)FPS
      17.028M16.053.731.735.6
      27.262M19.354.832.424.7
      36.911M19.156.134.325.6
      46.583M19.356.935.127.1
      56.942M21.456.334.627.1
      66.982M21.656.835.226.9
      77.013M23.457.135.226.5
      86.686M23.658.335.926.9
      97.875M26.859.737.122.1
    Tools

    Get Citation

    Copy Citation Text

    Yunzuo ZHANG, Cunyu WU, Yameng LIU, Tian ZHANG, Yuxin ZHENG. Joint self-attention and branch sampling for object detection on drone imagery[J]. Optics and Precision Engineering, 2023, 31(18): 2723

    Download Citation

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

    Category: Information Sciences

    Received: Dec. 29, 2022

    Accepted: --

    Published Online: Oct. 12, 2023

    The Author Email: ZHANG Yunzuo (zhangyunzuo888@sina.com)

    DOI:10.37188/OPE.20233118.2723

    Topics