Denoising atomic resolution 4D scanning transmission electron microscopy data with tensor singular value...

Area of Science:

• Materials Science
• Data Science
• Microscopy

Background:

• Atomic-resolution 4D STEM generates large datasets.
• Denoising is crucial for extracting meaningful information from these datasets.
• Existing denoising methods can be computationally intensive.

Purpose of the Study:

• To apply Tensor Singular Value Decomposition (SVD) for denoising 4D STEM data.
• To evaluate the performance and efficiency of Tensor SVD compared to other methods.
• To demonstrate the utility of Tensor SVD on both simulated and experimental microscopy data.

Main Methods:

• Tensor SVD was applied to simulated and experimental 4D STEM datasets.
• Performance was assessed using peak signal-to-noise ratio (PSNR).
• Processing times were compared against conventional denoising techniques.

Main Results:

• Tensor SVD achieved an average PSNR of ~40 dB on simulated data.
• Processing times were over 100 times faster than other methods.
• Experimental datasets (SrTiO3, LiZnSb/GaSb) were denoised in minutes on a standard PC.
• Improved quality of convergent beam electron diffraction patterns and annular dark field images.

Conclusions:

• Tensor SVD is a highly efficient and effective method for denoising 4D STEM data.
• This technique significantly reduces computational time without compromising data quality.
• Tensor SVD enhances the analysis of advanced materials using electron microscopy.