Related Experiment Video
Updated: Dec 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Breadth-first piston diagnosing approach for segmented mirrors through supervised learning of multiple-wavelength
Applied Optics
|November 11, 2020
Summary
This study introduces a novel piston diagnosing method for segmented mirrors using breadth-first search and machine learning. It accurately measures piston error without 2π ambiguity, regardless of submirror configuration.
Area of Science:
- Optical Engineering
- Machine Learning Applications
- Computational Optics
Background:
- Machine learning piston diagnosing for segmented mirrors is successful but faces 2π ambiguity.
- Existing methods' accuracy is often limited by submirror location and quantity.
Purpose of the Study:
- To develop a piston diagnosing approach for segmented mirrors that overcomes 2π ambiguity.
- To create a method whose accuracy is independent of submirror arrangement and count.
Main Methods:
- Utilized breadth-first search (BFS) algorithm combined with supervised learning on multi-wavelength images.
- Generated an object-independent, normalized dataset from in-focal and defocused images at various wavelengths.
- Employed deep convolutional neural networks for predicting piston aberration ranges and values after segmenting mirrors into binary tree sub-models traversed by BFS.
Main Results:
- Achieved high accuracy in piston error measurement, with Pearson correlation coefficients exceeding 0.99.
- Demonstrated an average root mean square error of approximately 0.01λ for segmented mirrors.
- Confirmed accuracy independence from the location and number of submirrors.
Conclusions:
- The developed technique effectively measures piston error in segmented mirrors without 2π ambiguity.
- The approach is adaptable to real-world collected data and applicable to segmented mirrors with varying numbers of submirrors.
