The potential for Bayesian compressive sensing to significantly reduce electron dose in high-resolution STEM images

Area of Science:

• Materials Science
• Electron Microscopy
• Computational Imaging

Background:

• High-resolution imaging in scanning transmission electron microscopy (STEM) is often limited by sample sensitivity to electron beams, causing damage.
• This beam damage is a significant issue in fields like organic systems study, tomography, and in situ experiments.

Purpose of the Study:

• To demonstrate that alternative image acquisition strategies can mitigate electron beam damage in STEM.
• To apply compressive sensing techniques for reduced-dose, high-resolution STEM imaging.

Main Methods:

• Utilized compressive sensing via Bayesian dictionary learning on high-resolution STEM images.
• Applied computational algorithms to image datasets with a reduced number of sampled pixels, down to 5% of the original.

Main Results:

• Successfully recovered original high-resolution STEM images from datasets with only 5% of the original pixels.
• Validated the method's effectiveness for both atomic-resolution and nanometer-resolution imaging, including tomography datasets.
• Demonstrated applicability on strontium titanate and zeolite samples.

Conclusions:

• Compressive sensing with Bayesian dictionary learning offers a viable low-dose acquisition strategy for STEM.
• This computational approach can be integrated into existing STEM workflows without altering microscope hardware.
• Enables high-resolution imaging with reduced electron beam exposure, preserving delicate samples.