Zhang Haobo, Yang Yanrong, Zhang Zitao, Yin Chun, Wang Shengqian, Wei Kai, Chen Hao, Zhao Junlei
National Laboratory on Adaptive Optics, Chengdu 610209, China.
School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China.
Biomed Opt Express. 2024 Oct 25;15(11):6531-6548. doi: 10.1364/BOE.541483. eCollection 2024 Nov 1.
Shack-Hartmann-based wavefront sensing combined with deep learning, due to its fast, accurate, and large dynamic range, has been widely studied in many fields including ocular aberration measurement. Problems such as noise and corneal reflection affect the accuracy of detection in practical measuring ocular aberration systems. This paper establishes a framework comprising of a noise-added model, Hartmannograms with corneal reflections and the corneal reflection elimination algorithm. Therefore, a more realistic data set is obtained, enabling the convolutional neural network to learn more comprehensive features and carry out real machine verification. The results show that the proposed method has excellent measurement accuracy. The root mean square error (RMSE) of the residual wavefront is 0.00924 ± 0.0207 (mean ± standard deviation) in simulation and 0.0496 ± 0.0156 in a real machine. Compared with other methods, this network combined with the proposed corneal reflection elimination algorithm is more accurate, speedier, and more widely applicable in the noise and corneal reflection situations, making it a promising tool for ocular aberration measurement.
基于夏克-哈特曼的波前传感技术与深度学习相结合，因其具有快速、准确且动态范围大的特点，已在包括眼像差测量在内的许多领域得到广泛研究。在实际的眼像差测量系统中，噪声和角膜反射等问题会影响检测的准确性。本文建立了一个由加噪模型、带有角膜反射的哈特曼图以及角膜反射消除算法组成的框架。因此，获得了一个更真实的数据集，使卷积神经网络能够学习更全面的特征并进行真机验证。结果表明，所提出的方法具有出色的测量精度。在模拟中，残余波前的均方根误差（RMSE）为0.00924±0.0207（均值±标准差），在真机中为0.0496±0.0156。与其他方法相比，该网络结合所提出的角膜反射消除算法在噪声和角膜反射情况下更准确、更快且应用更广泛，使其成为眼像差测量的一个有前途的工具。
