Aberration Determination Using Pseudo-PSF and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional sensorless adaptive optics require multiple exposures or images to correct aberrations in optical systems, which is inefficient and can be challenging in high-resolution microscopy where objects are three-dimensional.
Innovation Solution
A method involving obtaining two images with different adaptive optical element configurations, applying a transform to these images to produce transformed images, calculating a ratio of the transformed images to determine a pseudo-PSF, and using machine learning to extract aberration coefficients from the pseudo-PSF data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional sensorless adaptive optics methods are used to correct aberrations, then aberration correction is achieved, but multiple exposures or images are required which reduces efficiency
Solution Approach 1:
The patent segments the aberration correction process into two distinct phases: (1) acquiring a pair of images with different known aberrations applied, and (2) using machine learning to directly predict aberration coefficients from these images. This segmentation allows the system to eliminate the need for multiple sequential exposures while maintaining correction accuracy, as the ML model can infer aberrations from just two images.
Solution Approach 2:
The patent replaces the conventional iterative optimization algorithm (mechanical/search-based approach) with a machine learning model that directly predicts aberration coefficients. This substitution eliminates the need for multiple trial exposures and optimization iterations, achieving both high accuracy and efficiency in a single prediction step after initial training.
2Measurement precision
If multiple exposures are used to correct aberrations, then accurate aberration estimation is achieved, but the process becomes time-consuming and complex
Solution Approach 1:
The patent performs preliminary training of the machine learning model using a dataset of images with known aberrations and their corresponding correction outcomes. This preliminary action allows the model to learn the relationship between images and aberration coefficients in advance, so that during actual operation, aberration estimation can be performed rapidly without requiring multiple sequential exposures or iterative corrections.
Solution Approach 2:
The patent uses a pair of images with different known aberrations as a simplified copy or representation of the full aberration state. Instead of requiring multiple exposures to sample different aberration conditions, the system captures just two representative images that contain sufficient information for the ML model to infer the complete aberration profile, significantly reducing measurement time.
3Reliability
If conventional methods require at least 2N+1 measurements to correct N modes, then complete aberration correction is achieved, but device complexity and operational complexity increase
Solution Approach 1:
The patent changes the fundamental parameter from the number of measurements required to the type of information extracted from measurements. Instead of requiring 2N+1 measurements with conventional optimization algorithms, the system uses a machine learning model that extracts aberration coefficients directly from just two images by learning the mapping between image characteristics and aberration parameters, thereby simplifying the measurement sequence while maintaining complete correction capability.
Data Source
AI summary
A method of determining aberration in an optical system comprising an adaptive optical element is provided. The method comprises obtaining a first image in which the adaptive optical element is in a first configuration. The method comprises obtaining a second image in which the adaptive optical element is in a second configuration, wherein the second configuration is different from the first configuration. The method comprises applying a transform to the first image and the second image to produce a transformed first image and a transformed second image. The method comprises obtaining a ratio comprising the transformed first image and the transformed second image and determining a pseudo-PSF from the ratio. The method comprises providing data sampled from the pseudo-PSF to a machine learning algorithm that has been trained to determine an output indicative of aberration coefficients from the data.