raplima/petrog_thin_section_cnn

Python scripts used to perform petrographyc thin section analysis using convolutional neural networks

Python and R scripts used to perform petrographyc thin section classification using convolutional neural networks as was published in:

Petrographic microfacies classification with deep convolutional neural networks

Journal title: Computers and Geosciences
Authors: Rafael Pires de Lima, David Duarte, Charles Nicholson, Roger Slatt, Kurt J. Marfurt
Corresponding author: Rafael Pires de Lima

@article{
  PIRESDELIMA2020104481,
  title = "Petrographic microfacies classification with deep convolutional neural networks",
  journal = "Computers & Geosciences",
  volume = "142",
  pages = "104481",
  year = "2020",
  issn = "0098-3004",
  doi = "https://doi.org/10.1016/j.cageo.2020.104481",
  url = "http://www.sciencedirect.com/science/article/pii/S0098300419307629",
  author = "Rafael {Pires de Lima} and David Duarte and Charles Nicholson and Roger Slatt and Kurt J. Marfurt",
  keywords = "Petrography thin section analysis, Rock thin section, Convolutional neural networks, Transfer learning",
}

Data

Data can be downloaded here.

Scripts

• simple_cb.py: performs color balancing. **Please note the original source and references for the script and algorithm. **
• cnn_figs.py: creates (completely or partially) figures used for the paper
• cnn_processing.py: functions to fine tune CNNs
• data_manipulation.py: data preparation
• cnn_evaluate.py: used to evaluate of CNN models generated with cnn_processing.py
• test_pub_data.py: functions for the evaluation of public data
• metrics_and_confusion_matrix_plot.R: R files for metrics and confusion plots
• metrics_and_confusion_matrix_plot_for_public.R: R files for metrics and confusion plots

For an easier to use tool for initial transfer learning evaluation, an user iterface is provided here.

Software here is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.