[Paper Review] Deep Learning Segmentation of Complex Features in Atomic-Resolution Phase Contrast Transmission Electron Microscopy Images

Robbie Sadre, Colin Ophus+2 more|arXiv (Cornell University)|Dec 9, 2020

Advanced Electron Microscopy Techniques and Applications61 references36 citations

TL;DR

This paper proposes a U-Net-based deep learning model for automated segmentation of surface contaminants in atomic-resolution phase-contrast TEM images of graphene, outperforming conventional Bragg filtering. The method achieves higher accuracy and robustness with minimal hyperparameter tuning, enabling fully automated analysis of complex defect structures in 2D materials.

ABSTRACT

Phase contrast transmission electron microscopy (TEM) is a powerful tool for imaging the local atomic structure of materials. TEM has been used heavily in studies of defect structures of 2D materials such as monolayer graphene due to its high dose efficiency. However, phase contrast imaging can produce complex nonlinear contrast, even for weakly-scattering samples. It is therefore difficult to develop fully-automated analysis routines for phase contrast TEM studies using conventional image processing tools. For automated analysis of large sample regions of graphene, one of the key problems is segmentation between the structure of interest and unwanted structures such as surface contaminant layers. In this study, we compare the performance of a conventional Bragg filtering method to a deep learning routine based on the U-Net architecture. We show that the deep learning method is more general, simpler to apply in practice, and produces more accurate and robust results than the conventional algorithm. We provide easily-adaptable source code for all results in this paper, and discuss potential applications for deep learning in fully-automated TEM image analysis.

Motivation & Objective

• To address the challenge of automated segmentation of surface contaminants in atomic-resolution phase-contrast TEM images of graphene.
• To compare the performance of deep learning (U-Net) against conventional Bragg filtering for contaminant segmentation.
• To develop a generalizable, robust, and easily adaptable deep learning pipeline for automated TEM image analysis in materials science.
• To enable fully automated downstream analysis of defect structures in 2D materials by eliminating manual segmentation bottlenecks.

Proposed method

• A U-Net convolutional neural network architecture is trained on a dataset of labeled HRTEM images of graphene with surface contaminants.
• The model uses k-fold cross-validation to ensure robust generalization and avoid overfitting with limited labeled data.
• Image preprocessing includes normalization and intensity correction to account for beam-induced intensity variations across the field of view.
• The U-Net employs skip connections between encoder and decoder blocks to preserve spatial resolution and improve segmentation accuracy.
• Model training uses the Adam optimizer with binary cross-entropy loss for pixel-wise classification of contaminant vs. graphene regions.
• The method is compared against Bragg filtering, a classical Fourier-based technique relying on periodicity and diffraction patterns.

Experimental results

Research questions

• RQ1 Can a deep learning model outperform conventional Bragg filtering in segmenting surface contaminants in atomic-resolution TEM images of graphene?
• RQ2 How robust and generalizable is the U-Net model across diverse graphene microstructures and contamination patterns?
• RQ3 To what extent can the U-Net model reduce manual intervention in automated TEM image analysis pipelines?
• RQ4 How does the model perform on limited labeled datasets typical in scientific imaging applications?

Key findings

• The U-Net model achieves significantly higher segmentation accuracy than Bragg filtering, particularly in regions with complex, non-periodic contamination patterns.
• The deep learning approach is more robust to variations in electron dose and beam-induced intensity gradients across the image.
• The U-Net model generalizes well across different graphene grain boundaries and contamination morphologies without retraining.
• The method reduces the need for manual segmentation, enabling fully automated analysis pipelines for large-scale TEM datasets.
• The authors provide open-source code for the U-Net model and training pipeline, facilitating reuse and extension by the scientific community.
• The U-Net approach enables reliable detection of contaminant layers that obscure atomic-scale defect features, such as grain boundaries.