/Simple-CNN-ThinSection-XPLPPL

Simple convolutional neural network model to classify between XPL and PPL thin sections

Primary LanguageJupyter Notebook

Simple CNN Thin Section XPL or PPL Classifier

Trained model (contained in model folder) where a pre-trained .h5 keras model (trained using 2180 images) is used for distinguishing between PPL and XPL thin sections

Example of result given in the notebook (.ipynb)

Classified XPL heart

Result of classification on XPL and PPL in test images

Clasification of all images in testimgs folder

Abour the model elaboration:

The model was trained on 2180 images (1090 XPL, 1090 PPL), due to a lack of computational power I reduced the image quality from around 1000x1000 px to 400x400px to train the model. All images are RGB.

The Training dataset and valitadion dataset were splitted 80% - 20% and data augmentation was used (width shift, height shift, zoom, flip).

Accuracy on training set: 94%

Accuracy on validation set: 88%

Model Used

Standard model for binary classification with data augmentation:

Standard model used

Training and Validation Accuracy

Matplotlib graph with train and val acc

Maybe should have continued with more epochs...

Training and Validation Loss

Matplotlib graph with train and val loss

-- UNAL, Iván Ferreira