Laser & Optoelectronics Progress, Volume. 60, Issue 24, 2428005(2023)

Dual-Stream Feature Aggregation Network for Unmanned Aerial Vehicle Aerial Images Semantic Segmentation

Runzeng Li1, Zaifeng Shi1,3、*, Fanning Kong1, Xiangyang Zhao1, and Tao Luo2
Author Affiliations
  • 1School of Microelectronics, Tianjin University, Tianjin 300072, China
  • 2College of Intelligence and Computing, Tianjin University, Tianjin 300072, China
  • 3Tianjin Key Laboratory of Imaging and Sensing Microelectronic Technology, Tianjin 300072, China
  • show less
    Figures & Tables(8)
    Network architecture and partial module. (a) Overall network architecture; (b) ConvNeXt block; (c) coordinate attention block
    BGA module
    Comparison of prediction maps of different models on AeroScapes dataset. (a) Picture 002001_049; (b) picture 038032_032; (c) picture 045002_049; (d) picture 310019_016; (e) picture 311000_004
    Comparison of prediction maps of different models on Semantic Drone dataset. (a) Picture 002; (b) picture 056; (c) picture 119; (d) picture 311; (e) picture 412
    • Table 1. Comparison of evaluation results of different models on AeroScapes dataset

      View table

      Table 1. Comparison of evaluation results of different models on AeroScapes dataset

      MethodBackbonemIoU /%mPA /%
      FCN3VGG-162967.5974.53
      U-Net4ResNet-502075.8483.31
      PSPNet7MobileNetV33058.1563.86
      PSPNet7ResNet-502060.5766.72
      RefineNet831ResNet-1012063.0970.82
      DeepLabV3+10MobileNetV33078.0184.3
      DeepLabV3+10Xception3277.4985.03
      DADA2731DeepLabV23381.5388.75
      DSRL28,[31ResNet-1012082.4889.72
      ProposedXception3283.1690.75
    • Table 2. Comparison of evaluation results using coordinate attention block at different locations of CA-ASPP

      View table

      Table 2. Comparison of evaluation results using coordinate attention block at different locations of CA-ASPP

      LocationmIoU /%mPA /%
      479.3587.29
      579.4187.53
      679.5587.67
      4,679.7187.90
      1,2,3,478.8687.68
      1,2,3,4,679.6287.88
    • Table 3. Evaluation results of ablation experiments with different improved methods

      View table

      Table 3. Evaluation results of ablation experiments with different improved methods

      MethodmIoU /%mPA /%
      -77.4985.03
      CA-ASPP79.7187.90
      ConvBranch78.9686.84
      BGAModule79.4587.38
      ConvBranch,BGAModule80.8588.04
      CA-ASPP + ConvBranch + BGAModule81.5388.96
      ConvBranch + BGAModule + Multi-loss82.8490.31
      CA-ASPP + ConvBranch + BGAModule + Multi-loss83.1690.75
    • Table 4. Comparison of evaluation results of different models on Semantic Drone dataset

      View table

      Table 4. Comparison of evaluation results of different models on Semantic Drone dataset

      MethodBackbonemIoU /%mPA /%
      FCN3VGG-162954.6163.63
      U-Net4ResNet-502057.3868.45
      PSPNet7MobileNetV33045.4354.08
      PSPNet7ResNet-502042.8151.55
      DeepLabV3+10MobileNetV33055.3164.56
      DeepLabV3+10Xception3255.4864.00
      ProposedXception3272.0980.34
    Tools

    Get Citation

    Copy Citation Text

    Runzeng Li, Zaifeng Shi, Fanning Kong, Xiangyang Zhao, Tao Luo. Dual-Stream Feature Aggregation Network for Unmanned Aerial Vehicle Aerial Images Semantic Segmentation[J]. Laser & Optoelectronics Progress, 2023, 60(24): 2428005

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Remote Sensing and Sensors

    Received: Mar. 27, 2023

    Accepted: Apr. 23, 2023

    Published Online: Nov. 27, 2023

    The Author Email: Shi Zaifeng (shizaifeng@tju.edu.cn)

    DOI:10.3788/LOP230955

    Topics