轨道角动量(OAM)复用和编码技术可有效提高光通信系统信道容量。近些年研究者提出将机器学习(ML)技术用于OAM 模式探测以提高OAM 光通信系统性能。本文对基于机器学习的OAM 模式探测方案进行了综述，包括误差反向传播(BP)神经网络、自组织神经网络(SOM)、支持向量机(SVM)、卷积神经网络(CNN)、光束变换辅助的识别技术以及全光衍射深度神经网络(D2NN)，分析了各类机器学习OAM 探测器在对抗大气、水下信道带来的干扰时展现出的性能差异以及各自优势。

The orbital angular momentum (OAM) multiplexing and encoding technologies can effectively increase the channel capacity of the optical communication systems. In recent years, some researchers focus on using machine learning (ML) technology to detect OAM modes to improve the performance of OAM optical communication system. In this paper, the OAM modes detecting schemes based on ML technology are reviewed, including error back-propagating (BP) neural networks, self-organizing feature map (SOM), support vector machine (SVM), convolutional neural network (CNN), mode recognition techniques base on beam transformations and all-optics diffractive deep neural networks (D2NN). The performance, advantages and obstacles of each kind of the neural networks in atmosphere and underwater channels are analyzed.