Physics Informed Neural Network Enhanced Denoising for Atomic Resolution STEM Imaging

TL;DR

This paper introduces a Physics-Informed Neural Network framework that effectively denoises atomic resolution STEM images by preserving structural details and physical signal integrity, improving material analysis capabilities.

Contribution

The novel PINN-based denoising method integrates spectral fidelity, total variation, and brightness/contrast losses to enhance atomic resolution STEM images.

Findings

Effective noise reduction while preserving atomic details

Maintains physical signal intensities and structural integrity

Complementary to existing denoising methods

Abstract

Atomic resolution STEM images often suffer from noise due to low electron doses and instrument imperfections, hence it is challenging to obtain critical structural details required for material analysis. To address the problem, we propose a Physics-Informed Neural Network (PINN) framework for denoising STEM images. Our method integrates spectral fidelity, total variation, and brightness/contrast consistency losses to ensure the preservation of fine structures, smooth regions, and physical signal intensities, maintaining the structural integrity of the denoised images. Our proposed method effectively balances noise reduction with the preservation of atomic resolution details and complements existing methods, seeking to enhance the utility of STEM images in material characterization and analysis.