• Laser & Optoelectronics Progress
  • Vol. 60, Issue 2, 0228001 (2023)
Kuo Zhang1, Zhangjin Chen1,2,*, Dong Qiao1, and Yan Zhang1
Author Affiliations
  • 1Microelectronics Research and Development Center, Shanghai University, Shanghai 200444, China
  • 2Modern Educational Technology Center, Shanghai University, Shanghai 200444, China
  • show less
    DOI: 10.3788/LOP212698 Cite this Article Set citation alerts
    Kuo Zhang, Zhangjin Chen, Dong Qiao, Yan Zhang. Real-Time Image Detection via Remote Sensing Based on Receptive Field and Feature Enhancement[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0228001 Copy Citation Text show less
    YOLOv4 network structure diagram
    Fig. 1. YOLOv4 network structure diagram
    Bottleneck structure
    Fig. 2. Bottleneck structure
    Structure design of PANet-lite
    Fig. 3. Structure design of PANet-lite
    Receptive field enhancement module
    Fig. 4. Receptive field enhancement module
    Channel attention mechanism structure
    Fig. 5. Channel attention mechanism structure
    Spatial attention mechanism structure
    Fig. 6. Spatial attention mechanism structure
    Improved network structure
    Fig. 7. Improved network structure
    Training loss curve of proposed model
    Fig. 8. Training loss curve of proposed model
    Detecting results comparison
    Fig. 9. Detecting results comparison
    ModuleTChannelNumStrideOutputName
    Conv2D3212208×208×32
    Bottleneck11611208×208×16
    Bottleneck62422104×104×24
    Bottleneck6323252×52×32M1
    Bottleneck6644252×52×64
    Bottleneck6963126×26×96M2
    Bottleneck61603213×13×160
    Bottleneck63201113×13×320M3
    Table 1. Description of MobileNetV2 structure after pruning
    ParameterBaselineSEECACBAMRFBRFB+CBAM
    mAP /%85.3086.4687.3188.2786.4289.80
    Weight /MB41.6441.8341.6642.0343.2243.61
    Table 2. Ablation experiment comparison results
    ModelBackboneInput sizemAP /%Weight /MBDetection speed /(frame·s-1
    Faster-RCNNResNet50600×60088.161114.8
    SSDVGG16300×30074.799525.2
    YOLOv3DarkNet-53416×41685.0824120.2
    YOLOv4CSPDarkNet-53416×41691.2825122.5
    YOLOv4-TinyCSPDarkNet-Tiny416×41680.752335.2
    LM2-YOLOv4MobileNetV2416×41689.804433.4
    Table 3. Comparison of experimental results between proposed model and other models
    Kuo Zhang, Zhangjin Chen, Dong Qiao, Yan Zhang. Real-Time Image Detection via Remote Sensing Based on Receptive Field and Feature Enhancement[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0228001
    Download Citation