STEM Diffraction Pattern Analysis with Deep Learning Networks

TL;DR

The paper tackles the bottleneck of grain orientation mapping from STEM diffraction patterns by training DL models to predict Euler angles directly from 4D SPED data. It compares CNN, DenseNet-121, and Swin Transformer architectures, finding that attention-based Swin Transformer models provide the best overall accuracy and coherence in resulting crystal maps, especially for overlapping grains. A systematic hyperparameter study highlights the importance of data quality, appropriate batch size, and learning-rate scheduling for stable convergence. The approach enables automated, high-resolution orientation mapping and holds promise for accelerated microstructure analysis in energy materials, with future work aimed at using simulated labels and accounting for thickness effects.

Abstract

Accurate grain orientation mapping is essential for understanding and optimizing the performance of polycrystalline materials, particularly in energy-related applications. Lithium nickel oxide (LiNiO$_{2}$) is a promising cathode material for next-generation lithium-ion batteries, and its electrochemical behaviour is closely linked to microstructural features such as grain size and crystallographic orientations. Traditional orientation mapping methods--such as manual indexing, template matching (TM), or Hough transform-based techniques--are often slow and noise-sensitive when handling complex or overlapping patterns, creating a bottleneck in large-scale microstructural analysis. This work presents a machine learning-based approach for predicting Euler angles directly from scanning transmission electron microscopy (STEM) diffraction patterns (DPs). This enables the automated generation of high-resolution crystal orientation maps, facilitating the analysis of internal microstructures at the nanoscale. Three deep learning architectures--convolutional neural networks (CNNs), Dense Convolutional Networks (DenseNets), and Shifted Windows (Swin) Transformers--are evaluated, using an experimentally acquired dataset labelled via a commercial TM algorithm. While the CNN model serves as a baseline, both DenseNets and Swin Transformers demonstrate superior performance, with the Swin Transformer achieving the highest evaluation scores and the most consistent microstructural predictions. The resulting crystal maps exhibit clear grain boundary delineation and coherent intra-grain orientation distributions, underscoring the potential of attention-based architectures for analyzing diffraction-based image data. These findings highlight the promise of combining advanced machine learning models with STEM data for robust, high-throughput microstructural characterization.

Figures (17)

• Figure 1: Training data ($DS_{Train}$) and evaluation data ($DS_{Val}$ and $DS_{Test}$) acquisition via DP and TM label classification.
• Figure 2: Workflow for the microstructure investigation.
• Figure 3: General training structure of trained ML models.
• Figure 4: Structure of trained CNN models.
• Figure 5: Structure of trained DenseNets models, including Denselayer and Transitionlayer structure.