Mineral-Identification

This repo provides the dataset and code for reproducing the experiments in paper ---- Mineral Identification Based on Deep Learning That Combines Image and Mohs Hardness.

Data

• We crawled 36 species of mineral images from the mindat website, a total of 183,688 images.
• The specific species are as follows: Agate, Albite, Almandine, Anglesite, Azurite, Beryl, Cassiterite, Chalcopyrite, Cinnabar, Copper, Demantoid, Diopside, Elbaite, Epidote, Fluorite, Galena, Gold, Halite, Hematite, Magnetite, Malachite, Marcasite, Opal, Orpiment, Pyrite, Quartz, Rhodochrosite, Ruby, Sapphire, Schorl, Sphalerite, Stibnite, Sulphur, Topaz, Torbernite, Wulfenite.

How to run code

Dependency

• pytorch

Run

• If you have only one GPU, you can run EfficientNet_36classes_trainABatch_ValABatch.py and EfficientNet_hardness_36classes_trainABatch_ValABatch.py directly.

  # in bash
  python EfficientNet_36classes_trainABatch_ValABatch.py

  # in bash
  python EfficientNet_hardness_36classes_trainABatch_ValABatch.py

• If you have more than one GPU, you can use distributed training.

  # in bash
  # nproc_per_node is your GPU nums
  python -m torch.distributed.launch --nproc_per_node=2 EfficientNet_36classes_trainABatch_ValABatch.py

  # in bash
  # nproc_per_node is your GPU nums
  python -m torch.distributed.launch --nproc_per_node=2 EfficientNet_hardness_36classes_trainABatch_ValABatch.py

Application

• An app on smartphones using our trained model is implemented.

• For the Android or ios version, you can search "mineral identification" in Google Play or Apple's app store. And you can also visit link Android version and ios version directly.