An opensource library dedicated for analysis RHEED patterns and aim to extract information for phase mapping and crystal growth monitoring using machine learning algorithm.
H. Liang, V. Stanev, A. G. Kusne, Y. Tsukahara, K. Ito, R. Takahashi, M. Lippmaa, and I. Takeuchi, Application of Machine Learning to Reflection High-Energy Electron Diffraction Images for Automated Structural Phase Mapping, Phys. Rev. Mater. 6, 63805 (2022).
Mannul installation is required for now:
git clone https://github.com/AuroraLHT/rhana.git
cd rhana
pip install -e .Dependencies should be resolved by pip automatically. Check the pytorch website on how to install a proper version if you have issues related with fastai or pytorch.
The idea of this package is to provide tools to quickly analysis the RHEED pattern while the experiment is done or even still running. The method described in the paper is demonstrated in several notebooks which illustrate how the information is extracted in step by step manner. The method works on RHEED video or even just RHEED of before and after deposition. The workflow is seperated into multiple components and is highly controlable. Users could use their observation/intuition to guide or even swap the output of one of the component. Despite the nature of RHEED is extremely complex, we wish to extract as much of structural information as possible for a totally unknown system. These information could be linked with other structural characterization technique to quickly realize the phase composition of novel materials and faciliate the connection between structure and properties.
The U-Net model could be downloaded from here and should be placed in the learner folder. To train your own model or reproduce the learner with the demo asset, please refer to the Fastaiv2UNetTraining Notebook.
| Name | Description | What you would get |
|---|---|---|
| ApplyUNetwithTracking | Use U-Net to generate feature masks for every rheed patterns and also center and crop out the region of interest | Masks in run-line encoding (RLE) that is exported to a csv file |
| DynamicPhaseChange | Observe how the pattern evolve over time | A plot of variation in time of the relative intensity of each periodicity (data not provided in demo asset) |
| Fastaiv2UNetTraining | Train an U-Net model to predict masks for spots and streaks | A UNet model |
| MaterialMatching | Match materials from extracted periodicity | Phases that could be used to explain the RHEED pattern (heavily depend on experty in Materials Science) |
| PhaseMap-PhaseAnalysis | Extract the coexisting phases among a combi experiment with RHEED | All periodicity could be found from a combi experiemnt and the "phase" composition of each sample computed from the relative intensity of each periodicity |
Any general image labeling software would allow you to create mask for each image. Masks of each individual features could be stored also depending on the software implementation. The software used in this paper is django-labeller which is a user friendly open-source software.