• Opto-Electronic Engineering
  • Vol. 49, Issue 6, 210409 (2022)
Chonghui Zheng1、2, Tianshu Wang1、2、*, Zheqi Liu1、2, Qiaochu Yang1、2, and Xianzhu Liu1、2
Author Affiliations
  • 1National and Local Joint Engineering Research Center of Space Optoelectronics Technology, Changchun University of Science and Technology, Changchun, Jilin 130022, China
  • 2College of Opto-Electronic Engineering, Changchun University of Science and Technology, Changchun, Jilin 130022, China
  • show less
    DOI: 10.12086/oee.2022.210409 Cite this Article
    Chonghui Zheng, Tianshu Wang, Zheqi Liu, Qiaochu Yang, Xianzhu Liu. Deep transfer learning method to identify orbital angular momentum beams[J]. Opto-Electronic Engineering, 2022, 49(6): 210409 Copy Citation Text show less
    LG beams with topological charges of +2 and +5. (a) Topological charge is +2; (b) Topological charge is +5
    Fig. 1. LG beams with topological charges of +2 and +5. (a) Topological charge is +2; (b) Topological charge is +5
    The mapping relationship between LG beam and data
    Fig. 2. The mapping relationship between LG beam and data
    Phase screen of r0=0.24 m and r0=0.09 m and normalized phase screen.(a) Phase screen of r0=0.24 m; (b) Phase screen of r0=0.09 m; (c) Normalized phase screen of r0=0.24 m; (d) Normalized phase screen of r0=0.09 m
    Fig. 3. Phase screen of r0=0.24 m and r0=0.09 m and normalized phase screen.

    (a) Phase screen of r0=0.24 m; (b) Phase screen of r0=0.09 m;

    (c) Normalized phase screen of r0=0.24 m; (d) Normalized phase screen of r0=0.09 m

    Transfer learning architecture based on VGG16
    Fig. 4. Transfer learning architecture based on VGG16
    System structure
    Fig. 5. System structure
    Intensity diagram of ±5 order LG beam when D/r0 is 1.5 and 4. (a) D/r0=1.5; (b) D/r0=4
    Fig. 6. Intensity diagram of ±5 order LG beam when D/r0 is 1.5 and 4. (a) D/r0=1.5; (b) D/r0=4
    Training accuracy and validation accuracy of different recognition models. (a) Based on transfer learning; (b) Based on VGG16
    Fig. 7. Training accuracy and validation accuracy of different recognition models. (a) Based on transfer learning; (b) Based on VGG16
    Confusion matrix of transfer learning model in different turbulent environment. (a) D/r0=1.5 ; (b) D/r0=4
    Fig. 8. Confusion matrix of transfer learning model in different turbulent environment. (a) D/r0=1.5 ; (b) D/r0=4
    Confusion matrix of VGG16 model in different turbulent environment. (a) D/r0=1.5; (b) D/r0=4
    Fig. 9. Confusion matrix of VGG16 model in different turbulent environment. (a) D/r0=1.5; (b) D/r0=4
    IndicatorsTransfer learning modelVGG16 model
    Highest validation accuracy/%96.5698.25
    Total parameters6426896134301520
    Total training time/s1388732022
    Average epoch training time/s694.351601.1
    Table 1. Model evaluation indicators
    Chonghui Zheng, Tianshu Wang, Zheqi Liu, Qiaochu Yang, Xianzhu Liu. Deep transfer learning method to identify orbital angular momentum beams[J]. Opto-Electronic Engineering, 2022, 49(6): 210409
    Download Citation