Home

Several attempts were carried out to check if a small CNN model, based on the LeNet-5 model could classify rock images. 239 rock thin-section images including three rock types ”metamorphic”, ”sedimentary”, ”volcanic” were selected from a dataset, wich contained 2568 images, found on an online science database (Science Data Bank). The model has been trained and validated, after that the performances were evaluated with Accuracy:

Loss:

Confusion Matrix:

The model revealed a good performance notwithstanding the database shortness.