• Laser & Optoelectronics Progress
  • Vol. 59, Issue 24, 2415005 (2022)
Yu Zhang, Jie Ma*, Jinwen Cui, Yuehua Zhao, and Hong Liu
Author Affiliations
  • School of Electronics and Information Engineering, Hebei University of Technology, Tianjin, 300401, China
  • show less
    DOI: 10.3788/LOP202259.2415005 Cite this Article Set citation alerts
    Yu Zhang, Jie Ma, Jinwen Cui, Yuehua Zhao, Hong Liu. Rotation Target Detection Algorithm for Remote Sensing Image Using Attention Mechanism[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2415005 Copy Citation Text show less
    YOLOv5 network structure
    Fig. 1. YOLOv5 network structure
    Long edge definition method
    Fig. 2. Long edge definition method
    PSANeck structure
    Fig. 3. PSANeck structure
    PSA module structure
    Fig. 4. PSA module structure
    ECALayer structure
    Fig. 5. ECALayer structure
    YOLOv5 PAN structure
    Fig. 6. YOLOv5 PAN structure
    Improved feature fusion structure
    Fig. 7. Improved feature fusion structure
    Densely coded label
    Fig. 8. Densely coded label
    Improved network structure diagram
    Fig. 9. Improved network structure diagram
    YOLOv5m multi-scale characteristic thermodynamic diagram. (a) High resolution branch; (b) medium resolution branch; (c) low resolution branch
    Fig. 10. YOLOv5m multi-scale characteristic thermodynamic diagram. (a) High resolution branch; (b) medium resolution branch; (c) low resolution branch
    Improved algorithm multiscale characteristic thermodynamic diagram. (a) High resolution branch; (b) medium resolution branch; (c) low resolution branch
    Fig. 11. Improved algorithm multiscale characteristic thermodynamic diagram. (a) High resolution branch; (b) medium resolution branch; (c) low resolution branch
    Comparison of detection effect between YOLOv5m and improved algorithm. (a)(b)(c) detection results of YOLOv5m algorithm; (d) (e) (f) detection results of improved algorithm
    Fig. 12. Comparison of detection effect between YOLOv5m and improved algorithm. (a)(b)(c) detection results of YOLOv5m algorithm; (d) (e) (f) detection results of improved algorithm
    ωPLSHSVLVmAP
    180/886.5586.9073.5876.8280.96
    180/3287.4587.3376.0077.4082.04
    180/6487.4687.3476.0080.2082.75
    180/12887.4487.3276.0180.1882.73
    180/18087.3987.3376.0479.9082.67
    Table 1. Comparison of model performance under different angle discretization granularity ω
    GroupPSAECABiFPNDCLPLSHSVLVGFlopsmAP
    G1××××78.7979.4664.6161.2353.771.02
    G2×××86.5379.4764.6069.3449.874.99
    G3××87.4579.6071.5969.6449.877.07
    G4×86.5486.8969.9476.8252.980.05
    G587.4687.3476.0080.2052.982.75
    Table 2. Comparison of ablation experiments of each improved module
    MethodPLSHSVLVmAP
    FR-O2379.4237.1635.3038.0247.48
    IE-Net2480.2052.5849.7165.0161.88
    SCRDet89.9872.4168.3660.3272.77
    RSDet2590.1073.6070.2078.7078.15
    R-YOLOv5m87.4687.3476.0080.2082.75
    Table 3. Comparison of results of different algorithms on DOTA dataset subset
    MethodR2CNN26R2PN27Gliding Vertex28R-YOLOv5m
    mAP /%73.0779.6088.2088.89
    Table 4. Comparison of different algorithms on HRSC2016 ship dataset
    Yu Zhang, Jie Ma, Jinwen Cui, Yuehua Zhao, Hong Liu. Rotation Target Detection Algorithm for Remote Sensing Image Using Attention Mechanism[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2415005
    Download Citation