A task-oriented framework for efficient lithological mapping of imbalanced categories using hyperspectral imagery
Hyperspectral lithological mapping is a critical task in geological surveys and mineral resource exploration. However, the distribution of geological units is typically imbalanced and heterogeneous, with practical tasks requiring high-efficiency mapping over large regions. Traditional methods, such...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843225003966 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Hyperspectral lithological mapping is a critical task in geological surveys and mineral resource exploration. However, the distribution of geological units is typically imbalanced and heterogeneous, with practical tasks requiring high-efficiency mapping over large regions. Traditional methods, such as patch-based and regular window approaches, struggle to effectively address these challenges. In this study, we propose a novel fast sample-based window lithological mapping (FSWLM) framework that integrates three key innovations: (1) a sample-based window strategy to reduce computational redundancy and improve efficiency, (2) an TFM (Threshold and Frequency Modulation)-based sampling strategy to address class imbalance during training, and (3) an encoder-decoder network characterized by feature-spectrum fusion (FSF) pattern. FSWLM offers an integrated solution tailored to the geological demands of lithological mapping, optimizing both efficiency and accuracy. Experimental validation on the Dajiling dataset, a geologically complex region in Tibet, demonstrates that FSWLM outperforms state-of-the-art methods in classification accuracy, minor class recognition, and spatial coherence. These results underscore the effectiveness of the task-oriented framework in addressing the unique challenges of hyperspectral lithological mapping, with promising implications for large-scale geological surveys and mineral resource exploration. |
|---|---|
| ISSN: | 1569-8432 |