Geological Object Recognition in Geological Maps Through Data Augmentation and Transfer Learning Techniques

This study proposes an innovative method to improve geological object recognition by leveraging legend data for data augmentation and using transfer learning with EfficientNet. The approach enhances model performance, particularly for texture-rich datasets, by increasing data diversity and reducing training time, leading to more accurate and efficient geological feature classification.

folder

/data/input data after text removal

/data/input data after data overlays and data augmentation

/data/ov.fit Digital Elevation Model (DEM) data used in the manuscript

/or_data original (raw) data

/model saved model

files

EfficientNet.py EfficientNet model

data_augmentation.py Data augmentation processing

image_classification.py A fully connected layer was added to the model to adapt it for geological feature classification

image_classification_evaluate.py Evaluation of model results