Orbital Angular Momentum Modes Recognition Method Based on Transfer Learning
During the transmission of vortex beams in free-space optical communication channels, atmospheric turbulence causes wavefront phase disturbances, making identifying orbital angular momentum (OAM) modes more difficult. Additionally, the model requires extensive training datasets, which significantly...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Photonics Journal |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10933501/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | During the transmission of vortex beams in free-space optical communication channels, atmospheric turbulence causes wavefront phase disturbances, making identifying orbital angular momentum (OAM) modes more difficult. Additionally, the model requires extensive training datasets, which significantly prolongs the training time. In this paper, a method combining transfer learning with an improved convolutional neural network (CNN) is proposed to identify the OAM modes of distorted vortex beams. Compared to methods without transfer learning, this approach maintains higher accuracy while significantly reducing the training time and computational resources required. Furthermore, the impact of the model on OAM modes recognition accuracy is analyzed under varying atmospheric turbulence intensities and propagation distances, with comparisons made against traditional CNN models and the CNN models of other literature. To verify the recognition performance of the proposed model under adverse weather conditions, the impact of varying rainfall intensity and fog concentration on recognition accuracy was analyzed. In addition, the generalization ability of the model is evaluated after training on both single and mixed datasets. The results demonstrate that the recognition accuracy of the proposed method surpasses that of the traditional CNN models and the CNN models of other literature, achieving a 95.6% accuracy after 3 km of propagation under strong turbulence. Under strong turbulence, heavy rainfall, and heavy fog, the recognition accuracy at 1 km is 89.4% and 88.7%, respectively. Additionally, the model trained on the mixed dataset achieves a recognition accuracy of 97.08%, indicating robust generalization capability. The method proposed in this paper achieves high recognition accuracy while effectively reducing computational resources and training time, which is of great significance to the research on OAM modes recognition. |
|---|---|
| ISSN: | 1943-0655 |