Deep-learning-based Location Recognition Method for Detecting Multi-beam Multi-OAM Vortex Waves

Orbital angular momentum (OAM) has attracted attention since it could provide infinite orthogonal channels and effectively improve spectrum utilization in free-space optical communications. Although OAM pattern recognition widely utilizes deep learning techniques, the advanced identification (equipped with precise localization) of multi-beams remains challenging. In response to this challenge, we propose a new method that employs transfer learning based on a convolutional neural network (CNN). The approach features the capabilities of localization and recognition of various vortex beams. We train the model using phase distribution images of OAM beams of 9 types, rather than the amplitude distributions. Through simulation and experimentation, we demonstrate that the proposed method reduces the training time and improves the detection accuracy by up to 98%. These results contribute to the demultiplexing system of free-space optical OAM systems.