Identification of Crystal Symmetry from Noisy Diffraction Patterns by A Shape Analysis and Deep Learning

Introduction

This work contributes a novel descriptor Shaped Diffraction Pattern (Shaped DP) and Multi-stream DenseNet (MSDN) for crystal structure classification.

We also provide the codes as follows:

1. Shaped DP (see Shaped descriptor)
2. Pre-trained model of MSDN (using 72 Space Groups) (see MSDN)

Compatibility

We tested the codes with:

1. Tensorflow-GPU 1.13.1 under Ubuntu 18.04 and Anaconda3 (Python 3.7)
2. Tensorflow-GPU 1.13.1/Tensorflow 1.13.1 under Windows 10 and Anaconda3 (Python 3.7)

Requirements

1. Anaconda3
2. TensorFlow-GPU 1.13.1 or Tensorflow 1.13.1
3. PIL
4. NatSort
5. SciPy
6. Matplotlib
7. Numpy 1.16.4

Sample code

Shaped DP

• Run the code test_descriptor.py in Shaped descriptor

$ python test_descriptor.py

Pre-trained model: MSDN

• Run the code test_MSDN.py in MSDN

$ python test_MSDN.py --batch_size 16 --plot_sample 0

Dataset

The dataset is shared on Zenodo open data repository.

Citation

Please cite us if you are using our model or dataset in your research works:

[1] Leslie Ching Ow Tiong, Jeongrae Kim, Sang Soo Han and Donghun Kim, "Identification of crystal symmetry from noisy diffraction patterns by a shape analysis and deep learning," npj Computational Materials, 6:196, 2020. (See link).