Optical clock networks have distinct advantages for the dissemination of time/frequency, geodesy, and fundamental research. To realize such a network, the telecom band and optical atomic clocks have to be coherently bridged. Since the telecom band and optical atomic clocks reside in a distinct spectral region, second-harmonic generation is usually introduced to bridge the large frequency gap. In this paper, we introduce a new method to coherently link a 1550 nm continuous wave laser with a Ti:sapphire mode-locked laser-based optical frequency comb. By coupling the 1550 nm continuous wave laser light and the Ti:sapphire comb light together into a photonic crystal fiber, nonlinear interaction takes place, and new comblike frequency components related to the 1550 nm laser frequency are generated in the visible region. Consequently, we can detect beat notes between two combs in the visible region with a signal-to-noise ratio of more than 40 dB in a resolution bandwidth of 300 kHz. With this signal, we realize an optical frequency divider for converting the frequency of optical clocks in the visible region to the telecom band at 1.55 μm. An out-of-loop measurement shows that the additional noise and uncertainty induced in optical frequency conversion are 5×10-18 at 1 s averaging time and 2.2×10-19, respectively, which are limited by the uncompensated light path fluctuation but fulfill precision measurement using state-of-the-art optical clocks.

为了保证航空相机的成像效果，设计了气密光学窗口及其保护罩。本文对光学窗口的设计流程进行了分析，讨论了光学窗口设计的考虑因素，例如光学材料、光学厚度、压力和气密安装方案等。为了实现在非工作状态下对光学窗口的保护，设计了双层舱门结构形式的光窗保护罩，实现了系统的小型化。通过仿真分析和试验，测试了气密光学窗口和光窗保护罩的稳定性，系统能够在2个大气压和高低温环境下正常工作，性能良好。光学窗口保护罩整个厚度仅为37 mm，能够实现快速开启与关闭，单程时间为7.7 s。相关分析方法解决了光学窗口设计问题的盲目性，使光学窗口的设计更加合理可靠。本文提出的方案为航空相机光学窗口的设计和保护提供了参考和技术支持。

由于视网膜病变的类间图像特征差别小及分类临界值相对模糊，自动分级算法存在识别与分级准确率低的问题，提出了融合高效通道注意力（Efficient Channel Attention， ECA）特征的ConvNeXt视网膜病变自动分级模型。针对数据集中数据不足的问题，采用水平翻转左右变换的方法扩充数据，并引入相关数据集来均衡数据的分布。针对眼底图像中出现的图像模糊、光照不均等问题，采用Graham方法对图像进行预处理突出病变特征。提出了融合注意力的ConvNeXt网络来辅助医生诊断视网膜病变，引入ECA机制，并设计了E-Block模块，该模块具有高性能、低参数的特性，能够在训练过程中有效捕捉跨通道交互的信息，同时避免降维。采用迁移学习方法训练网络的所有层参数，加入dropout方法避免ConvNeXt网络的学习能力过强导致的过拟合问题。实验结果表明，所提出的模型敏感性为95.20%，特异度为98.80%，准确率为95.21%。与常用的网络相比，本文方法针对视网膜病变自动分级各项性能指标均有提高。

We present the frequency control of a 759 nm laser as a lattice laser for an ytterbium (Yb) optical clock. The frequency stability and accuracy are transferred from the Yb optical clock via an optical frequency comb. Although the comb is frequency-stabilized on a rubidium microwave clock, the frequency instability of the 759 nm laser is evaluated at the 10-15 level at 1 s averaging time. The frequency of the 759 nm laser is controlled with an uncertainty within 1 Hz by referencing to the Yb clock transition. Such a frequency-controlled 759 nm laser is suitable for Yb optical clocks as the lattice laser. The technique of laser frequency control can be applied to other lasers in optical clocks.

We report two ultra-stable laser systems automatically frequency-stabilized to two high-finesse optical cavities. By employing analog-digital hybrid proportional integral derivative (PID) controllers, we keep the merits of wide servo bandwidth and servo accuracy by using analog circuits for the PID controller, and, at the same time, we realize automatic laser frequency locking by introducing digital logic into the PID controller. The lasers can be automatically frequency-stabilized to their reference cavities, and it can be relocked in 0.3 s when interruption happens, i.e., blocking and unblocking the laser light. These automatic frequency-stabilized lasers are measured to have a frequency instability of 6×10-16 at 1 s averaging time and a most probable linewidth of 0.3 Hz. The laser systems were tested for continuous operation over 11 days. Such ultra-stable laser systems in long-term robust operation will be beneficial to the applications of optical atomic clocks and precision measurement based on frequency-stabilized lasers.