This repository is a toolbox for the easy training and application of convolutional neural networks for the segmentation of transmission electron microscopy images of carbon black agglomerates.

The utilized convolutional neural network was inspired by the Panoptic FPN architecture, developed by Kirillov et al. and is based on an implementation of Yakubovskiy, realized with PyTorch.

• CarbonBlackSegmentation
• Table of Contents
• Examples
• Citation
• Setup
• Getting started

Demos demonstrating the use of this codebase are provided in the demos folder of this repository. Classifying a set of images generally makes use of wrapper functions, which take a single or set of image paths and output a series of binary masks identifying agglomerates in the image. For example,

import os
from wrapper import single_image # segmenter for single image

# Specify image path.
image_path = os.path.join("test_images", "201805A_A6_004.png")

# Classify the default image
print("Classifying image...")
mask = single_image(image_path)
print("Complete.")

will read in a single test image, 201805A_A6_004.png, and output a mask, identifying pixels that are part of the agglomerates as 1 and pixels that are part of the background as 0.

Several example classifications are shown below.

If you use this repository for a publication, then please cite it using the following bibtex-entry:

@article{Sipkens.2021,
    author = {Timothy A. Sipkens, Max Frei, Alberto Baldelli, Patrick Kirchen, Frank E. Kruis and Steven N. Rogak},
    year = {2021},
    title = {Characterizing soot in {TEM} images using a convolutional neural network},
    url = {https://doi.org/10.1016/j.powtec.2021.04.026}
    doi = {10.1016/j.powtec.2021.04.026}
}

1. Install conda.
2. Open a command line.
3. Clone this repository: git@github.com:maxfrei750/CarbonBlackSegmentation.git
4. Change into the folder of the repository: cd CarbonBlackSegmentation
5. Create a new conda environment: conda env create --file environment.yml
6. Activate the new conda environment: activate CarbonBlackSegmentation

If you just want to segment images, then please refer to the scripts in the demo folder.

If you want to improve your segmentation results by training a model on your own data, then please refer to train_model.py.