• Laser & Optoelectronics Progress
  • Vol. 59, Issue 8, 0810006 (2022)
Rongqi Jiang1、2、*, Zecong Ye1、2, Yueping Peng2、**, Guorong Xie1、2, and Heng Du3
Author Affiliations
  • 1Graduate Team, Engineering University of PAP, Xi'an , Shaanxi 710086, China
  • 2School of Information Engineering, Engineering University of PAP, Xi'an , Shaanxi 710086, China
  • 3School of Civil Engineering, Xinjiang University, Urumqi , Xinjiang 830000, China
  • show less
    DOI: 10.3788/LOP202259.0810006 Cite this Article Set citation alerts
    Rongqi Jiang, Zecong Ye, Yueping Peng, Guorong Xie, Heng Du. Lightweight Target Detection Algorithm for Small and Weak Drone Targets[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0810006 Copy Citation Text show less
    Partial pictures of datasets. (a) Dataset A; (b) Dataset B
    Fig. 1. Partial pictures of datasets. (a) Dataset A; (b) Dataset B
    Analysis of size of drone targets in two datasets. (a) Dataset A; (b) Dataset B
    Fig. 2. Analysis of size of drone targets in two datasets. (a) Dataset A; (b) Dataset B
    Structure diagram of YOLOv4-tiny algorithm. (a) YOLOv4-tiny; (b) CSPBlock
    Fig. 3. Structure diagram of YOLOv4-tiny algorithm. (a) YOLOv4-tiny; (b) CSPBlock
    Structure diagram of DTD-YOLOv4-tiny model
    Fig. 4. Structure diagram of DTD-YOLOv4-tiny model
    ShuffleNetV2 and improved backbone network structure. (a) ShuffleV2Block; (b) backbone network of ShuffleNetV2; (c) backbone network of proposed algorithm
    Fig. 5. ShuffleNetV2 and improved backbone network structure. (a) ShuffleV2Block; (b) backbone network of ShuffleNetV2; (c) backbone network of proposed algorithm
    FPN structure comparison of different detection models. (a) YOLOv4-tiny; (b) YOLOv4-tiny (YOLO-Head enhancement); (c) DTD-YOLOv4-tiny
    Fig. 6. FPN structure comparison of different detection models. (a) YOLOv4-tiny; (b) YOLOv4-tiny (YOLO-Head enhancement); (c) DTD-YOLOv4-tiny
    Working principle of reorg_layer
    Fig. 7. Working principle of reorg_layer
    Working principles of sub-pixel Conv and sub-pixel. (a) Sub-pixel Conv; (b) sub-pixel
    Fig. 8. Working principles of sub-pixel Conv and sub-pixel. (a) Sub-pixel Conv; (b) sub-pixel
    Comparison of accuracy and detection speed of different target detection models under different datasets. (a) Dataset A; (b) Dataset B
    Fig. 9. Comparison of accuracy and detection speed of different target detection models under different datasets. (a) Dataset A; (b) Dataset B
    Comparison of partial detection results of test set on different datasets. (a) YOLOv4-tiny (Dataset A); (b) DTD-YOLOv4-tiny (Dataset A); (c) YOLOv4-tiny (Dataset B); (d) DTD-YOLOv4-tiny (Dataset B)
    Fig. 10. Comparison of partial detection results of test set on different datasets. (a) YOLOv4-tiny (Dataset A); (b) DTD-YOLOv4-tiny (Dataset A); (c) YOLOv4-tiny (Dataset B); (d) DTD-YOLOv4-tiny (Dataset B)
    AlgorithmImage sizemAP@50 /%GFLOPsDetection speed /(frame·s-1
    YOLOv441665.760.155
    YOLOv3-tiny41633.15.6345
    YOLOv4-tiny41640.26.9330
    Table 1. Performance comparison of different algorithms in MS COCO dataset
    ModelHead improveNeck improveBackbone ImproveAP /%ParametersGFLOPs
    Anchor box imporveYOLO-Head enhancement

    ShuffleV2

    Block(B)

    reorg

    layer

    sub

    pixel

    Shufflev2

    Block(B)

    ShuffleNetv2 BackboneProposed Backbone
    YOLOv4-tiny (Baseline)43.15.8×10612.1
    Imporve A84.15.8×10612.1
    Imporve B91.36.1×10614.3
    Imporve C93.66.8×10617.1
    Imporve D94.56.5×10616.1
    Imporve E87.10.65×1061.9
    Imporve F87.40.64×1062.2
    Imporve G85.30.27×1060.9

    DTD-

    YOLOv4-tiny

    89.40.32×1061.1
    Table 2. Ablation experiment results of DTD-YOLOv4-tiny algorithm
    AlgorithmImage sizeP /%R /%AP /%ParametersGFLOPsGPU /GB
    YOLOv3-SPP416×23493.596.296.8625×106116.93.85
    YOLOv4416×23485.696.695.5639×106105.95.6
    YOLOv4-tiny416×23481.587.684.15.87×10612.10.7
    DTD-YOLOv4-tiny416×23483.586.789.40.327×1061.10.52
    DTD-YOLOv4-tiny608×34288.393.994.20.327×1061.10.583
    DTD-YOLOv4-tiny960×54087.895.695.00.327×1061.11.3
    Table 3. 4 Performance comparison of different algorithms on Dataset A
    ModelImage sizeAP /%Model size
    YOLOv3181333×80084.9234.1 MB
    Fast RCNN(MobileNet)181333×80067.5162.5 MB
    Casacade RCNN(MobileNet)181333×80078.0384.9 MB
    TIB-Net181333×80089.2697.0 KB
    YOLOv4-tiny1344×75678.523 MB
    DTD-YOLOv4-tiny960×54080.31.4 MB
    DTD-YOLOv4-tiny1344×75683.31.4 MB
    DTD-YOLOv4-tiny1920×108085.11.4 MB
    Table 4. Performance comparison of different algorithms on Dataset B
    Rongqi Jiang, Zecong Ye, Yueping Peng, Guorong Xie, Heng Du. Lightweight Target Detection Algorithm for Small and Weak Drone Targets[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0810006
    Download Citation