LCC-Net: Swin transformer-CNN hybrid for enhanced land cover classification in natural disaster monitoring
Land cover classification (LCC) from satellite images is crucial in identifying and monitoring natural disasters, including cyclones, earthquakes, floods, and wildfires. Statistics reveal that accurate disaster classification from satellite data can enhance response times by up to 30 % and improve p...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Systems and Soft Computing |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772941925001218 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Land cover classification (LCC) from satellite images is crucial in identifying and monitoring natural disasters, including cyclones, earthquakes, floods, and wildfires. Statistics reveal that accurate disaster classification from satellite data can enhance response times by up to 30 % and improve prediction accuracy by approximately 25 %. However, existing methods need more accuracy due to varying image resolutions and difficulty distinguishing between similar land cover types under different disaster conditions. This research proposes a specialized network named the land cover classification network, referred to as LCC-Net, for classifying the land covers from satellite images. This method involves initial image normalization, noise reduction, and enhancing spatial resolution to improve classification performance. Here, an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) is implemented to reduce the noise with the super-resolution concept, improving image quality by combining deep learning with adversarial training. The core of LCC-Net employs the Swin Transformer Convolutional Neural Network (ST-CNN), which leverages self-attention mechanisms to capture intricate spatial features and temporal dynamics. The ST-CNN outperforms traditional CNN models by providing a better contextual understanding of land cover variations associated with different disaster scenarios. To further enhance classification accuracy, the Adaptive Moment Estimation (AME) optimizer is utilized for loss minimization, ensuring efficient convergence and improved model robustness. This approach aims to enhance the precision of disaster identification and response strategies across the four fundamental classes: Cyclone, Earthquake, Flood, and Wildfire. The LCC-Net achieved Accuracy (99.999 %), Precision (99.569 %), Recall (99.320 %), and F1-Score (99.270 %). Finally, LCC-Net delivers highly accurate image classification with remarkably fast processing speed, outperforming state-of-the-art approaches. |
|---|---|
| ISSN: | 2772-9419 |