Title: Machine learning meets Singular Optics II: Single-pixel Detection of Structured Light Show Chinese title

Title: 机器学习邂逅奇异光学 II：结构光的单像素检测

Abstract: Structured light beams, including Laguerre-Gaussian (LG), Hermite-Gaussian (HG), and perfect vortex (PV) spatial modes, have been at the forefront of modern optics due to their potential in communications, metrology, and sensing. Traditional recognition methods often demand complex alignments and high-resolution imaging. Speckle-learned recognition (SLR) has emerged as a powerful alternative, exploiting the spatio-temporal speckle fields generated by light-diffuser interactions. This paper builds upon the earlier report: Machine Learning Meets Singular Optics (Proc. SPIE 12938, 2024), which demonstrated structured light recognition using 2D speckle images in both on-axis and off-axis channels captured in the spatial domain. In the present work, the recognition framework is advanced by employing 1D speckle information captured in the spatial and temporal domains. This paper reviews how the 2D spatial information of the structured light is mapped on 1D speckle arrays captured in space and 1D temporal speckle fluctuations recorded in time. The 1D speckle arrays captured in the spatial domain have successfully recognised the parent structured light beams with accuracy exceeding 94%, even by employing 1/nth of the 2D speckle data. More recently, 2D spatial information of structured light beams has been mapped onto temporal speckle sequences recorded by a single-pixel detector in the temporal domain. This study highlights the accuracy exceeding 96% across various structured light families, with resilience to turbulence and modal degeneracy. These advances establish scalable, alignment-free, and low-latency recognition architectures suitable for optical communication, sensing, and quantum technologies. Show Chinese abstract

Abstract: 结构光束，包括拉盖尔-高斯（LG）、厄米-高斯（HG）和完美涡旋（PV）空间模式，由于其在通信、计量和传感方面的潜力，已成为现代光学的前沿。传统的识别方法通常需要复杂的对准和高分辨率成像。散斑学习识别（SLR）作为一种强大的替代方法，利用了光扩散器相互作用产生的时空散斑场。本文基于先前的报告：《机器学习遇见奇异光学》（Proc. SPIE 12938, 2024），该报告展示了使用在空间域中捕获的轴上和轴外通道中的二维散斑图像进行结构光识别。在本工作中，通过采用在空间和时间域中捕获的一维散斑信息，改进了识别框架。本文回顾了结构光的二维空间信息如何映射到空间中捕获的一维散斑阵列和时间中记录的一维时间散斑波动。在空间域中捕获的一维散斑阵列成功地识别了父结构光束，准确率超过94%，即使只使用了1/n的二维散斑数据。最近，结构光束的二维空间信息已被映射到时间域中由单像素探测器记录的时间散斑序列。这项研究突出了在各种结构光家族中准确率超过96%，并且对湍流和模式退化具有鲁棒性。这些进展建立了可扩展、无需对准和低延迟的识别架构，适用于光通信、传感和量子技术。