• Acta Photonica Sinica
  • Vol. 51, Issue 12, 1210003 (2022)
Hongjian FU1, Hongyang BAI1,*, Hongwei GUO1, Yuman YUAN1, and Weiwei QIN2
Author Affiliations
  • 1School of Energy and Power Engineering,Nanjing University of Science and Technology,Nanjing 210094,China
  • 2School of Nuclear Engineering,Rocket Force University of Engineering,Xi'an 710025,China
  • show less
    DOI: 10.3788/gzxb20225112.1210003 Cite this Article
    Hongjian FU, Hongyang BAI, Hongwei GUO, Yuman YUAN, Weiwei QIN. Object Detection Method of Optical Remote Sensing Image with Multi-attention Mechanism[J]. Acta Photonica Sinica, 2022, 51(12): 1210003 Copy Citation Text show less
    Schematic diagram of MA-YOLOv5 network structure
    Fig. 1. Schematic diagram of MA-YOLOv5 network structure
    Schematic diagram of ARFCA module(3 channels)
    Fig. 2. Schematic diagram of ARFCA module(3 channels)
    Schematic of the Swin transformer module
    Fig. 3. Schematic of the Swin transformer module
    The loss value of MA-YOLOV5 network during training and validation
    Fig. 4. The loss value of MA-YOLOV5 network during training and validation
    mAP values for each category of targets
    Fig. 5. mAP values for each category of targets
    Branch number of ARFCA

    Convolution kernel size

    of each branch

    mAPFPS
    15×568.050.9
    23×3,5×568.450.2
    33×3,5×5,7×768.549.4
    43×3,5×5,7×7,9×968.548.3
    Table 1. Detection accuracy and speed of ARFCA with different branch numbers
    MethodBackbonemAPAP50AP75APSAPMAPLFPS
    SSDResNet⁃5048.480.951.529.057.551.377
    RetinaNetResNet⁃5060.788.870.554.266.254.828
    FCOSResNet⁃5063.589.874.356.269.358.238
    YOLOv5CSPdarknet64.991.778.458.072.461.259
    YOLOV5⁃STRCSPdarknet66.392.680.258.673.362.952
    YOLOv5⁃ARFCACSPdarknet67.293.080.659.273.563.454
    MA⁃YOLOv5CSPdarknet68.593.482.860.376.565.349
    Table 2. The performance of different networks on the test set
    Hongjian FU, Hongyang BAI, Hongwei GUO, Yuman YUAN, Weiwei QIN. Object Detection Method of Optical Remote Sensing Image with Multi-attention Mechanism[J]. Acta Photonica Sinica, 2022, 51(12): 1210003
    Download Citation