• Laser & Optoelectronics Progress
  • Vol. 62, Issue 10, 1028003 (2025)
Lunming Qin1, Wenquan Mei1, Haoyang Cui1, Houqin Bian1,*, and Xi Wang2
Author Affiliations
  • 1College of Electronics and Information Engineering, Shanghai University of Electric Power, Shanghai 201306, China
  • 2School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing 100044, China
  • show less
    DOI: 10.3788/LOP241927 Cite this Article Set citation alerts
    Lunming Qin, Wenquan Mei, Haoyang Cui, Houqin Bian, Xi Wang. Improved YOLOv8s Object Detection Algorithm for Remote Sensing Image[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1028003 Copy Citation Text show less
    Structure of improved model network
    Fig. 1. Structure of improved model network
    Structures of BiFPN and EBiFPN networks. (a) BiFPN; (b) EBiFPN
    Fig. 2. Structures of BiFPN and EBiFPN networks. (a) BiFPN; (b) EBiFPN
    Structure of EMA attention mechanism
    Fig. 3. Structure of EMA attention mechanism
    Structure of EMABottleneck module
    Fig. 4. Structure of EMABottleneck module
    Structure of C2f-EB module
    Fig. 5. Structure of C2f-EB module
    Schematic diagram of SIoU parameters
    Fig. 6. Schematic diagram of SIoU parameters
    Label quantity of different datasets. (a) DOTA dataset; (b) RSOD dataset; (b) NWPU NHR-10 dataset
    Fig. 7. Label quantity of different datasets. (a) DOTA dataset; (b) RSOD dataset; (b) NWPU NHR-10 dataset
    Detection results of different algorithms on the NWPU NHR-10 dataset
    Fig. 8. Detection results of different algorithms on the NWPU NHR-10 dataset
    Comparison of detection results between YOLOv8 and proposed algorithm. (a)‒(c) YOLOv8; (d)‒(f) proposed algorithm
    Fig. 9. Comparison of detection results between YOLOv8 and proposed algorithm. (a)‒(c) YOLOv8; (d)‒(f) proposed algorithm
    ParameterConfiguration
    Operating systemUbuntu
    GPUNVIDIA GeForce RTX 3090
    Python3.8
    CUDA11.7
    Table 1. Experimental environment configuration
    NeckmAP50mAP50∶95
    FPN-PAN70.047.2
    BiFPN70.947.9
    EBiFPN71.749.1
    Table 2. Comparison of different feature fusion structures
    Loss functionmAP50mAP50∶95
    CIoU70.047.2
    WIoU70.847.7
    ShapeIoU70.047.7
    MPDIoU69.947.6
    SIoU70.347.8
    SWIoU71.548.3
    Table 3. Comparison of different loss functions
    MethodEBiFPNC2f-EBSWIoUmAP50 /%mAP50∶95 /%Params /MGFLOPs
    170.047.211.128.5
    271.749.112.231.8
    371.448.411.128.7
    471.548.311.128.5
    571.949.212.332.1
    673.049.512.332.1
    Table 4. Results of ablation experiments
    MethodmAP50 /%Params /MGFLOPsModel size /MB
    SSD47.324.0274.594.4
    Faster R-CNN50.4136.8401.8534.2
    CenterNet49.832.770.2127.9
    RetinaNet49.436.4146.3142.5
    YOLOv5m69.820.947.941.2
    YOLOXs67.69.026.835.2
    YOLOv7-Tiny66.56.013.012.0
    YOLOv8s70.011.128.522.0
    YOLOv8m72.225.979.350.8
    Proposed73.012.332.124.3
    Table 5. Detection results of different methods on the DOTA dataset
    MethodAP /%mAP50 /%Params /MGFLOPsModel size /MB
    AircraftOlitankOverpassPlayground
    SSD56.593.458.896.576.324.0274.594.4
    Faster R-CNN66.296.766.298.581.9136.8401.8534.2
    Cenernnet81.098.387.198.191.132.770.2127.9
    Retinanet75.998.787.198.390.036.4146.3142.5
    YOLOv5m92.094.288.893.992.220.947.941.2
    YOLOXs89.799.581.098.592.29.026.835.2
    YOLOv7-Tiny90.396.060.996.485.96.013.012.0
    YOLOv8s92.797.788.898.194.311.128.522.0
    YOLOv8m93.698.390.298.495.125.979.350.8
    Proposed93.996.593.998.695.712.332.124.3
    Table 6. Detection results of different methods on the RSOD dataset
    Lunming Qin, Wenquan Mei, Haoyang Cui, Houqin Bian, Xi Wang. Improved YOLOv8s Object Detection Algorithm for Remote Sensing Image[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1028003
    Download Citation