Multidimensional Diffractive Vortex Beam Recognition Based on Deep Learning

文章提出了一种基于深度学习算法的多维度衍射涡旋光束识别方法，旨在解决自由空间光通信(FSO)系统中因大气湍流和传播距离导致的涡旋光束(Vortex Beam)轨道角动量(OAM)模式的识别难题。涡旋光束因其携带的OAM模式具有正交性和无限扩展潜力，在光通信、量子计算和光镊系统等领域具有广泛应用。然而，传统OAM识别方法依赖精密光学设备，且对光路稳定性要求苛刻，难以应对湍流和传播距离带来的光束畸变和模式串扰。文章通过模拟涡旋光的衍射场，并结合深度学习算法，提出了一种能够同时识别涡旋光OAM模式和传播距离的方案。具体而言，利用空间光调制器(SLM)生成多束相干叠加的涡旋光，并通过改进的Non-Kolmogorov大气湍流模型和角谱衍射法模拟真实通信环境中的光束传播。通过引入Swin-Transformer深度学习模型，实现了对涡旋光OAM模式和传播距离的高效识别。实验结果表明，采用OAM模式集合(如{1, 2, 3}和{1, 2, 3, 4})扩展OAM复用类型的方法，相比单一OAM模式，在识别准确率和鲁棒性方面均有显著提升。在32类和64类分类任务中，实验组的验证集准确率分别达到99.4%和99.8%，显著优于对照组。该方法为自由空间OAM通信系统的解调提供了新的解决方案，并展示了深度学习在涡旋光束识别中的巨大潜力。本研究为克服涡旋光在湍流和传播距离影响下的OAM模式识别难题提供了新的思路，对推动自由空间光通信、量子通信和光学传感等领域的发展具有重要意义。This paper proposes a multidimensional diffraction vortex beam recognition method based on deep learning algorithms, aiming to address the challenges of Orbital Angular Momentum (OAM) mode recognition in vortex beams (Vortex Beam) caused by atmospheric turbulence and propagation distance in Free-Space Optical Communication (FSO) systems. Vortex beams, due to their orthogonal and infinitely scalable OAM modes, have broad applications in optical communications, quantum computing, and optical tweezer systems. However, traditional OAM recognition methods rely on precision optical equipment and stringent optical path stability, making them ill-suited to handle beam distortion and mode crosstalk induced by turbulence and propagation distance. In this paper, by simulating the diffraction field of vortex beams and incorporating deep learning algorithms, we propose a solution capable of simultaneously recognizing the OAM modes and propagation distances of vortex beams. Specifically, we generate multiple coherently superimposed vortex beams using a Spatial Light Modulator (SLM) and simulate beam propagation in real communication environments using an improved non-Kolmogorov atmospheric turbulence model and angular spectrum diffraction method. By introducing the Swin-Transformer deep learning model, we achieve efficient recognition of vortex beam OAM modes and propagation distances. Experimental results demonstrate that expanding OAM multiplexing types using OAM mode sets (such as {1, 2, 3} and {1, 2, 3, 4}) significantly improve recognition accuracy and robustness compared to single OAM modes. In 32-class and 64-class classification tasks, the validation set accuracy of the experimental group reaches 99.4% and 99.8%, respectively, significantly outperforming the control group. This method provides a new solution for demodulation in free-space OAM communication systems and demonstrates the immense potential of deep learning in vortex beam recognition. The research offers novel insights into overcoming the challenges of OAM mode recognition in vortex beams under the influence of turbulence and propagation distance, contributing significantly to the advancement of free-space optical communications, quantum communications, and optical sensing fields.