Geological object recognition in legacy maps through data augmentation and transfer learning techniques

Abstract

Maps are crucial tools in geosciences, providing detailed representations of the spatial distribution and relationships among geological features. Accurate recognition and classification of geological objects within these maps are essential for applications in resource exploration, environmental management, and geological hazard assessment. Along the years, many legacy geological maps have been accumulated, and many of them are not in data formats ready for machines to read and analyze. The inherent diversity and complexity of geological features, combined with the labor-intensive process of manual annotation, pose significant challenges in the usage of those maps. This study addresses these challenges by proposing an innovative approach that leverages legend data for data augmentation and employs transfer learning techniques to improve the quality of object recognition. Legend data from geological maps offer standardized symbols and annotations. Using them to augment existing datasets increases the diversity and volume of training data, thereby enhances the model's ability to generalize across various geological contexts. A deep learning model called EfficientNet is then fine-tuned using the augmented dataset to recognize and classify geological features more accurately. The model's performance is evaluated based on accuracy, recall, and F1-score, with results showing significant improvements, particularly for datasets with texture-rich information. The proposed method demonstrates that the combination of data augmentation and transfer learning significantly enhances the accuracy and efficiency of geological object recognition. This approach not only reduces the manual effort needed for geological object recognition but also contributes to the advancement of geological mapping and analysis.