Electron Microscopy Sparse Imaging via Beam Blanker and ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Electron microscopy faces challenges in capturing high-quality images due to electron beam-induced artifacts and incomplete data, which are time-consuming and costly to address, and existing methods like Compressive Sensing require expensive hardware modifications.
Innovation Solution
A system that intentionally under-samples data using a high-speed electrostatic beam blanker to reduce electron dose by up to 80% and employs dictionary learning algorithms like Beta Process Factor Analysis to reconstruct missing image portions, enabling simultaneous pattern recognition and image recovery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high energy electron flux is used to improve image quality, then image resolution is improved, but electron beam induced artifacts increase
Solution Approach 1:
The patent applies partial action by intentionally under-sampling the image data, capturing only a subset of the full image information. This reduces the electron dose by up to 80% while still enabling recovery of the complete image through machine learning algorithms. The system captures less data than traditionally required, demonstrating that full sampling is not necessary for high-quality reconstruction.
Solution Approach 2:
The patent replaces the traditional mechanical/optical image capture system with a machine learning-based reconstruction system. Instead of relying on complete physical sampling of the image, the system uses neural networks and generative models to synthesize the full image from partial measurements, substituting computational processing for physical data collection.
2Loss of information
If complete data is captured to ensure information completeness, then image quality is maintained, but time and cost increase
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models on large datasets of image information. This pre-learning enables the system to rapidly reconstruct complete images from partial data during actual operation, eliminating the need for time-consuming data recapturing. The computational model is prepared in advance to handle the reconstruction task efficiently.
Solution Approach 2:
The patent uses copying by generating synthetic complete images from partial measurements through machine learning. The system creates accurate replicas of the full image data based on learned patterns and relationships, rather than requiring physical capture of all original data. This computational copying preserves information completeness without the associated time and resource costs.
3Quantity of substance
If traditional Compressive Sensing is used to reduce sampling requirements, then data acquisition is reduced, but hardware modifications are required
Solution Approach 1:
The patent replaces hardware-based compressive sensing modifications with software-based machine learning reconstruction. Instead of modifying the electron microscope hardware to implement compressive sensing measurements, the system uses standard imaging hardware combined with advanced neural network algorithms to achieve the same data reduction and reconstruction goals, eliminating complex hardware changes.
Solution Approach 2:
The patent changes the approach from hardware parameter modifications to software parameter optimization. Rather than altering hardware configuration to enable compressive sensing, the system optimizes computational parameters such as network architecture, loss functions, and training data to achieve efficient reconstruction from reduced samples, maintaining hardware simplicity while achieving the desired data reduction.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach allows for high-quality image reconstruction at reduced electron doses, suppressing beam-induced artifacts and achieving similar or better image quality with multiple sparse images compared to one fully sampled image, while reducing noise and preserving sample integrity.
Implementation Method 1
A sparse imaging system, including a scanning electron microscope with a high-speed electrostatic beam blanker, is employed to directly reduce and suppress beam induced artifacts through under-sampling pixels
Implementation Method 2
The incomplete data and/or datasets are processed via one or more statistical machine learning algorithms that employ latent feature detection to 'in-paint' missing information into the incomplete data and/or datasets
Data Source
AI summary
Aspects of the present disclosure involve a data capturing and processing system that intentionally captures data and/or data sets with missing pieces of information. The data and/or datasets may include various types of data, such as one-dimensional signals, two-dimensional images (or other images), and/or three-dimensional structures. The captured data is processed to restore missing information into the data and/or data sets, thereby enabling simultaneous pattern recognition and image recovery.