Deep learning-based geological map generation using geological routes
Geological mapping has been used in several applications, including resource assessment, hazard prediction, and land-use planning. However, a substantial gap remains in the current state of research regarding methods for acquiring mapping object labels, recognizing mapping objects, and their practic...
Gespeichert in:
Veröffentlicht in: | Remote sensing of environment 2024-08, Vol.309, p.114214, Article 114214 |
---|---|
Hauptverfasser: | , , , , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | Geological mapping has been used in several applications, including resource assessment, hazard prediction, and land-use planning. However, a substantial gap remains in the current state of research regarding methods for acquiring mapping object labels, recognizing mapping objects, and their practical applications in geological mapping. To enhance the accuracy, research depth, and intelligence of geological mapping, while simultaneously reducing fieldwork demands and mitigating risks for geologists operating in challenging terrains, this study introduces an innovative framework based on deep learning to generate 1:50,000 scale regional geological maps. We utilized a deep neural network architecture with multiple shared feature fusion layers focused on geological routes. Compared with conventional methods and random forest-based lithological mapping approaches, this method excelled in harnessing the synergistic potential of multimodal data sources and extracted high-dimensional spatial features from cross-modal data, yielding a method particularly well-suited for geological mapping characterized by extensive data, fine-grained lithological classification within mapping units, limited samples, and data imbalance challenges. This method enabled the recognition of a diverse range of geological mapping objects and maintained an average recall rate of >90% for these classifications. Generating labels based on geological routes enhanced the practical applicability of this framework. Through rigorous experimentation across multiple 1:50,000 and 1:25,000 geological mapping scenarios, the generated artificial intelligence-driven geological maps closely approximated the level of authentic geological maps, with some mapping objects exhibiting superior distribution and precision compared with manually created maps. This approach is poised to transform traditional geological mapping methodologies and processes while advancing the intelligent application of geological surveys in mineral resource estimation, environmental assessments, and coastal zone mapping.
[Display omitted]
•A neural network-based geological map generation model was developed.•A high-confidence geological route label generation method was proposed.•Enhanced multimodal fusion feature extraction was achieved by the proposed model.•Intermediate fusion deep neural network performed better than early fusion.•AI geological maps offer a new way to improve the accuracy of geological mapping |
---|---|
ISSN: | 0034-4257 1879-0704 |
DOI: | 10.1016/j.rse.2024.114214 |