Deep Learning Phase Noise Reduction for Microstructure Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital holography techniques face challenges in accurately filtering out complex phase noise in micro-nano structure measurements due to the limitations of Gaussian noise models and the masking of phase distortion, leading to residual noise and reduced measurement accuracy.
Innovation Solution
A deep learning-based method using an end-to-end filtering convolutional neural network combined with a subspace projection method, which simulates noise using Brown and Perlin models to create mixed data sets for training, effectively filters out complex phase noise in digital holographic experiments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional phase filtering methods using Gaussian noise model are applied, then phase noise filtering is performed, but residual noise remains because the Gaussian model cannot simulate all noise sources
Solution Approach 1:
The patent creates a simulated training dataset by copying and combining multiple noise models (Gaussian, Brown, Perlin) to replicate complex real-world phase noise characteristics. This synthetic dataset is then used to train the neural network, enabling it to learn and filter all types of phase noise without requiring extensive experimental data collection.
Solution Approach 2:
The patent combines multiple noise models (Gaussian, Brown, Perlin) into a composite noise simulation framework. This composite approach allows the training data to encompass diverse noise characteristics, enabling the neural network to handle mixed noise sources that occur in actual digital holographic measurements.
2Measurement precision
If filtering is applied to wrapped phase map, then phase noise filtering is attempted, but phase noise characteristics are masked by phase distortion
Solution Approach 1:
The patent performs preliminary phase distortion compensation using Zernike polynomial fitting before the neural network filtering stage. By removing systematic phase distortions in advance, the characteristics of phase noise become more apparent and easier for the neural network to detect and filter, solving the masking problem.
3Measurement precision
If deep learning-based phase filtering methods are used, then phase noise filtering performance is improved, but network parameters increase and computation time increases
Solution Approach 1:
The patent performs preliminary dimensionality reduction by projecting the phase map onto a lower-dimensional subspace using Principal Component Analysis (PCA) before feeding it to the neural network. This preprocessing step reduces the input data complexity and network parameter requirements while preserving the essential noise characteristics needed for effective filtering.
Solution Approach 2:
The patent extracts only the essential noise-related features from the phase map by projecting onto a reduced subspace, separating the critical information from redundant data. This extraction process reduces computational complexity and network parameters while maintaining filtering effectiveness.
4Measurement precision
If deep learning-based phase filtering methods are used, then phase noise filtering performance is improved, but computation speed decreases
Solution Approach 1:
The patent performs preliminary subspace projection to reduce data dimensionality before neural network processing. By compressing the input phase map into a lower-dimensional representation, the computational burden is significantly reduced, enabling faster processing speeds while maintaining filtering performance.
Solution Approach 2:
The patent segments the filtering process into two distinct stages: (1) a fast subspace projection step that reduces dimensionality, and (2) a neural network filtering step that processes the compressed data. This segmentation allows the computationally intensive neural network to operate on smaller data, improving overall processing speed.
Data Source
AI summary
A deep learning-based digital holographic continuous phase noise reduction method for microstructure measurement is provided. A MEMS microstructure is simulated to generate an object phase image through generation of random matrix superposition, noise in a digital holographic continuous phase map is simultaneously simulated to generate a noise grayscale image, and a simulation data set is thus created. An end-to-end convolutional neural network is designed, and a trained convolutional neural network is trained and obtained. A holographic interference pattern of an object under measurement is collected by photographing, and after spectrum extraction, angular spectrum diffraction, phase unwrapping, and distortion compensation, a continuous phase map containing only the object phase and noise is obtained and input into the trained convolutional neural network to obtain an object phase map. A simulation data set is accurately created in the disclosure, thereby the difficulty of collecting a large amount of experimental data is avoided.