QDots-sizer

This is a python package for automated size classification of quantum dots (with average diameter within 10 nm) from x-ray scattering data, using a light all-convolutional deep learning model trained with physics-informed augmented data.

Supported data type

• Sample state: colloidal and dry
• Angular range: custom

Installation

This package

Just clone the current repository by typing in a terminal: git clone https://github.com/DeByeUSerSYstem/QDots-sizer.git

noisy

Please download this utility from https://github.com/SineBell/noisy and follow the relative installation guide.

Pre-requisites

In order to run the program, you need to have installed:

• Python (version 3.8 or higher) and pip (version 19.0 or higher) limits imposed by TensorFlow installation
• Python modules: numpy, pandas, matplotlib, scikit-learn
• TensorFlow: version 2.11.0 or higher (as backing for the Keras module) (*)

If needed, the modules named in the last two points can be installed via python -m pip install -U module_name

(*) Please note that both CPU and GPU installation are available (and supported), but have different installation procedures. Refer to the official TensorFlow documentation here.

Usage

Data Prediction

A pre-trained model for PbS is available in the repository. To classify any pattern of this material, go into the folder where you cloned the repository and run the following: python Predict_datum_by_name.py path_to_diffractogram -l wavelength The process should take just a few seconds. Both CPU and GPU can be used, depending on the TensorFlow version you have installed.

At the end of the procedure, you will find a new file.xy with the predicted size in the file name and containing your pattern as the the algorithm sees it. You will also be presented with the option of saving a plot of this pattern.

Argument
path_to_diffractogram: mandatory
Relative of Full path of the diffractogram you wish to have a size prediction on.
Data have to be in a two-columns xy file structure: the first column containing 2 angles, the second column the measured intensity.

Flags
-l wavelength: optional
Wavelength used during data collection, in
If omitted, the algorithm will use the K1-Cu ( = 1.5406 ) by default.

--header_rows N: optional
Number of header lines at the beginning of your data to be skipped.
If omitted, all lines will be read.
Any line beginning with # is automatically skipped by the program.

--help
Shows the automatic help topic.

Data Augmentation and Network Training

Two separate scripts can be found, one for each task. Within each script, one can find the various augmentation and training parameters, which can be modified by the user if needed. These parameters include, but are not limited to: nanocrystals/solvent ratio and noise levels, Q-range limits, training speed and duration, CNN architecture.

If you wish to create your own database and train your own model, just open a terminal, go to the folder where this repository has been cloned and type: python 1_augment_database.py path_to_your_simulation_folder && python 2_train_aCNN.py where path_to_your_simulation_folder is the path to the folder containing your initial diffractograms that will be augmented and then used as dataset for the model trainig. We recommend to simulate these initial diffractograms using the dedicated software Debussy, developed in our To.Sca.Lab group @University of Insubria (Italy).

If, instead, you wish to use the already augmented dataset we have made availble, then open a terminal and type: * cd data_training/: to go into the datatraining_ folder (a more detailed README can be found here, if you wish); * sh extract_files.sh: this will extract all the diffractogram, making them ready for the training sctipt to use; * cd ..: to go back one level, to the main project folder; * python 2_train_aCNN.py: to train the algorithm.

Attribution

Authors: Lucia Allara @goatworks, Federica Bertolotti @febertol, and Antonella Guagliardi.

Version: 1.0 / October, 2023

repo owner e-mail: lallara@uninsubria.it, lucia.allara@gmail.com
contributors e-mail: federica.bertolotti@uninsubria.it, antonella.guagliardi@ic.cnr.it

License: GPL-3.0 Please, acknowledge use of this work with the appropriate citation.

Citation: you can find it on the righ bar under the "Cite this repository" section.
Full paper OpenAccess: https://doi.org/10.1038/s41524-024-01241-6