Opt Lett. 2018 Sep 1;43(17):4264-4267. doi: 10.1364/OL.43.004264.
Most of the methods used today for the alignment of segmented mirrors are based on Shack-Hartman wavefront sensors. Other proposed methods are based on curvature sensors. These can be used to cross-check the measurements given by the primary method. We investigate a different approach which employs convolutional neural networks. This technique allows the piston step values between segments to be measured with high accuracy, as well as a large capture range at visible wavelengths. The technique does not require special hardware, and is fast to be used at any time during the observation.
目前用于分段反射镜对准的大多数方法都基于 Shack-Hartman 波前传感器。其他提出的方法基于曲率传感器。这些可以用来交叉检查主要方法给出的测量结果。我们研究了一种不同的方法，它采用卷积神经网络。该技术允许以高精度测量段之间的活塞步值，并且在可见光波长范围内具有较大的捕获范围。该技术不需要特殊的硬件，并且在观测过程中的任何时间都可以快速使用。
