Pan Feng, Dong Bin, Xiao Wen, Ferraro Pietro
Opt Express. 2020 Mar 2;28(5):6537-6551. doi: 10.1364/OE.387511.
Sub-aperture stitching in digital holography (DH) is a very important issue both for the spatial resolution improvement as well as for measuring larger aperture through synthetic enlargement of numerical aperture. In fact, sub-apertures stitching permits to greatly expand the capabilities of optical metrology thus allowing to accurately measure complex optical surfaces such as large spherical and aspheric. Stitching operations can be difficult and cumbersome depending on geometric parameters of specific objects under test. However, here we show that machine learning can definitively aid this process. In fact, here we propose for the first time, to the best of our knowledge, a novel sub-aperture stitching approach based on machine learning applied to an array of different phase-maps sub-apertures recorded by an off-axis digital holographic systems. Essentially, we construct a network according to computation model of sub-aperture stitching and remove the alignment errors and system aberration of sub-aperture maps by training the network. Correct measurement of the surface topography of hemisphere surface is demonstrated thus validating the proposed learning approach. Reported results demonstrate that machine learning can be a useful tool for simplifying the process and for making it a reliable and accurate tool in optical metrology.
数字全息术中的子孔径拼接对于提高空间分辨率以及通过数值孔径的合成扩大来测量更大孔径而言都是一个非常重要的问题。实际上，子孔径拼接能够极大地扩展光学计量的能力，从而可以精确测量诸如大型球面和非球面等复杂光学表面。根据特定被测物体的几何参数，拼接操作可能会困难且繁琐。然而，我们在此表明机器学习可以切实辅助这一过程。事实上，据我们所知，我们首次提出了一种基于机器学习的新型子孔径拼接方法，该方法应用于由离轴数字全息系统记录的一系列不同相位图子孔径。本质上，我们根据子孔径拼接的计算模型构建一个网络，并通过训练该网络来消除子孔径图的对准误差和系统像差。通过对半球面表面形貌的正确测量证明了所提出的学习方法的有效性。报告结果表明，机器学习可以成为简化该过程并使其在光学计量中成为可靠且准确工具的有用手段。
