Research on the Linearly Polarized Optical Mode Recognition Based on Optical Diffractive Neural Network

光模式识别在光通信、光学成像及传感等领域具有重要应用前景，尤其在提高系统容量与处理效率方面扮演关键角色。传统识别方法主要依赖于数字信号处理或衍射光学元件，存在计算复杂度高、功耗大等问题，或缺乏泛化处理能力。本文提出一种基于衍射光学神经网络(Diffractive Neural Networks, DNNs)的光学光模式识别框架，专注于线性偏振(LP)模式的识别任务。通过引入拉盖尔–高斯模式构建的大规模复振幅光场数据集(共2000组样本)，并设计四层衍射网络进行训练与优化，系统对模式的识别准确率达到94.6%。仿真结果显示本方案具备优异的识别精度，展示出良好的应用潜力。本文研究为下一代低功耗、高效率的全光学计算系统提供了理论依据与设计框架，也为未来物理实现奠定了基础。Optical mode recognition has significant application prospects in fields such as optical communication, optical imaging, and sensing, especially in enhancing system capacity and processing efficiency. Traditional recognition methods mainly rely on digital signal processing or diffractive optical elements, which have problems such as high computational complexity and high power consumption, or lack of generalization processing capabilities. This paper proposes an optical mode recognition framework based on diffractive neural networks (DNNs), focusing on the recognition task of linearly polarized (LP) modes. By introducing a large-scale complex amplitude optical field dataset constructed from Laguerre-Gaussian modes (with a total of 2000 samples), and designing a four-layer diffractive network for training and optimization, the system achieves a recognition accuracy of 94.6% for the modes. Simulation results show that this scheme has excellent recognition accuracy and demonstrates good application potential. This research provides a theoretical basis and design framework for next-generation low-power, high-efficiency all-optical computing systems, and lays the foundation for future physical implementation.