Zhang Haobo, Zhao Junlei, Chen Hao, Zhang Zitao, Yin Chun, Wang Shengqian
National Laboratory on Adaptive Optics, Chengdu 610209, China.
School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China.
Sensors (Basel). 2024 Apr 25;24(9):2728. doi: 10.3390/s24092728.
The Shack-Hartmann wavefront sensor (SHWFS) is widely utilized for ocular aberration measurement. However, large ocular aberrations caused by individual differences can easily make the spot move out of the range of the corresponding sub-aperture in SHWFS, rendering the traditional centroiding method ineffective. This study applied a novel convolutional neural network (CNN) model to wavefront sensing for large dynamic ocular aberration measurement. The simulation results demonstrate that, compared to the modal method, the dynamic range of our method for main low-order aberrations in ocular system is increased by 1.86 to 43.88 times in variety. Meanwhile, the proposed method also has the best measurement accuracy, and the statistical root mean square (RMS) of the residual wavefronts is 0.0082 ± 0.0185 λ (mean ± standard deviation). The proposed method generally has a higher accuracy while having a similar or even better dynamic range as compared to traditional large-dynamic schemes. On the other hand, compared with recently developed deep learning methods, the proposed method has a much larger dynamic range and better measurement accuracy.
夏克-哈特曼波前传感器（SHWFS）被广泛用于测量人眼像差。然而，个体差异导致的较大人眼像差容易使光斑移出SHWFS中相应子孔径的范围，从而使传统的质心算法失效。本研究将一种新型卷积神经网络（CNN）模型应用于波前传感，以测量大动态范围的人眼像差。仿真结果表明，与模态法相比，我们的方法在人眼系统中主要低阶像差的动态范围增加了1.86至43.88倍。同时，该方法还具有最佳的测量精度，残余波前的统计均方根（RMS）为0.0082±0.0185λ（均值±标准差）。与传统的大动态范围方案相比，该方法通常具有更高的精度，同时具有相似甚至更好的动态范围。另一方面，与最近开发的深度学习方法相比，该方法具有更大的动态范围和更好的测量精度。
