Microstructure Characterization II

This is the repository for microstructure characterization research II since May 2019. It is composed of two parts:

• Feature engineering for image classification
• Representation learning with GANs

Publication

This repo contains code for reproducing key results in Image driven machine learning based microstructure recognition and quantification on small datasets.

Our previous work: An image-driven machine learning approach to kinetic modeling of a discontinuous precipitation reaction.

Feature Engineering for Image Classification

Resources

Phase-specific Features

After segmentation, we extract area and shape information from

• train.py
• features
  • __init__.py
  • features.py
• classification
  • binary_classification.py

Binary Classification

Visualization

Area features

After features are extracted, you can plot the area features by running

python plot/area_features.py results/area_featurs.csv binary figures/area_features_binary.png

Run python plot/area_features.py -h for help. The supported output format are PNG (for static image output) and HTML (for interactive plot).

Visualization of Training Outputs

Plot the confusion matrix for binary classification

Before this step, make sure you have trained a binary classification model and have the confusion matrix results ready.

To plot the confusion matrix, run

python plot/confusion_matrix.py

The output figure will be saved to the ./figures directory.

Representation Learning with GANs

Resources

System requirements