• Opto-Electronic Engineering
  • Vol. 51, Issue 10, 240158 (2024)
Yuntang Li*, Wenkai Zhu, Hengjie Li, Juan Feng..., Yuan Chen, Jie Jin, Bingqing Wang and Xiaolu Li|Show fewer author(s)
Author Affiliations
  • College of Mechanical and Electrical Engineering,China Jiliang University,Hangzhou,Zhejiang 310018,China
  • show less
    DOI: 10.12086/oee.2024.240158 Cite this Article
    Yuntang Li, Wenkai Zhu, Hengjie Li, Juan Feng, Yuan Chen, Jie Jin, Bingqing Wang, Xiaolu Li. Image recognition of complex transmission lines based on lightweight encoder-decoder networks[J]. Opto-Electronic Engineering, 2024, 51(10): 240158 Copy Citation Text show less
    Structure of lightweight encoder-decoder network
    Fig. 1. Structure of lightweight encoder-decoder network
    Convolutional block attention module
    Fig. 2. Convolutional block attention module
    Bneck-CBAM module
    Fig. 3. Bneck-CBAM module
    Depth atrous spatial pyramid pooling module
    Fig. 4. Depth atrous spatial pyramid pooling module
    Visualization of feature maps at different depths
    Fig. 5. Visualization of feature maps at different depths
    Labelme labeling transmission lines
    Fig. 6. Labelme labeling transmission lines
    Network training loss value variation curves
    Fig. 7. Network training loss value variation curves
    Comparison of sparse training with different regularization coefficients
    Fig. 8. Comparison of sparse training with different regularization coefficients
    Comparison of four network recognition results for transmission lines
    Fig. 9. Comparison of four network recognition results for transmission lines
    训练参数数值
    Batch_size8
    Initial_lr0.0001
    Epoch500
    CudaTrue
    Table 1. Training parameters for lightweight encode-decoder network
    方法MPA/%MIoU/%FPS
    方法189.8783.2226
    方法291.4484.0224
    方法390.9283.8631
    方法492.3484.5729
    方法592.6784.6430
    Table 2. Results of ablation experiment
    λMPA/%MIoU/%Epoch
    092.6784.64500
    0.0191.5283.77500
    0.00192.0784.12500
    0.000192.5884.56500
    0.0000192.5684.55500
    Table 3. Comparison of sparse training experimental results with different regularization coefficients
    剪枝率MPA/%MIoU/%FPS参数量/(106)
    092.5884.56305.82
    0.192.3684.39325.27
    0.292.2484.28354.63
    0.392.1684.22374.12
    0.492.1184.19413.57
    0.591.7283.85442.92
    Table 4. Comparison of experimental results with different pruning rates
    网络MPA/%MIoU/%FPS参数量/(106)
    PSPNet[10]81.8673.789178
    U2Net[6]89.7282.31843.99
    文献[7]87.3779.622112.77
    轻量型编解码网络92.1184.19413.57
    Table 5. Comparison of four network recognition results
    Yuntang Li, Wenkai Zhu, Hengjie Li, Juan Feng, Yuan Chen, Jie Jin, Bingqing Wang, Xiaolu Li. Image recognition of complex transmission lines based on lightweight encoder-decoder networks[J]. Opto-Electronic Engineering, 2024, 51(10): 240158
    Download Citation