Peak area detection network (PADNet) directly learns to predict the phase region from the raw (2D) X-ray diffraction patterns (XRD) image without any need for explicit preprocessing and background removal. PADNet contains specially designed large symmetrical convolutional filters at the first layer to capture the peaks and automatically remove the background by computing the difference in intensity counts across different symmetries.
- Python 2.7 with Jupyter Notebook
- matplotlib 3.0.2
- scipy 1.2.0
- mpl_toolkits
- Pillow 5.1.0
- TensorFlow 1.12
- scikit-learn 0.20.2
This repository contains the code for preprocessing warped 2D X-ray diffraction patterns using minimum filter and convolutional smoothening, along with the code for training PADNet model on the XRD patterns.
bg_process.ipynb- code for preprocessing warped 2D XRD images using minimum filter and convolutional smoothening.load_data.py- code for loading XRD images for training PADNet model.model.py- implementation of the architecture and training of PADNet models using different datasets.perf_analysis.ipynb- Jupyter notebook to train PADNet models using different cross validation ratios on different datasets.train_utils.py- utility code for training PADNet model.training-data- folder containing instructions on how to obtain the dataset used in the paper.
Launch a jupyter notebook by using the following command in a terminal:
jupyter notebook
A notebook interface will appear in a new browser window or tab. Next open the perf_analysis.ipynb jupyter notebook file. It contains the code to run the PADNet model and also shows the expected output for each cell.
This notebook contains the complete output logs from all the models trained using different types of input data and evaluation using different types of test data performed in the paper [1]. The input for PADNet is composed of 2048x2048 warped XRD patterns. For SLAC model, the input has one XRD pattern with composition. For Bruker model, the input is composed of two XRD patterns along with the composition. The PADNet is architecture is defined for 8 phase region label classes, it can be modified to work with other XRD datasets with different number of classes by simple modification in the code. The PADNet model can be similarly trained and evaluated on other datasets.
Please cite the following work if you are using PADNet model and/or code for background preprocessing of 2D XRD patterns provided in this repository.
- Dipendra Jha, Aaron Gilad Kusne, Reda Al-Bahrani, Nam Nguyen, Wei-keng Liao, Alok Choudhary, and Ankit Agrawal, "Peak area detection network for directly learning phase regions from raw x-ray diffraction patterns." In 2019 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE. DOI:10.1109/IJCNN.2019.8852096
Copyright (C) 2019, Northwestern University.
See COPYRIGHT notice in top-level directory.
This work is supported in part by the following grants: NIST award 70NANB14H012, NSF award CCF-1409601; DOE awards DE-SC0014330, DE-SC0019358.