Pansharpening Toolkit¶
A comprehensive toolkit implementing both classic and state-of-the-art deep learning methods for fusing multispectral (MS) and panchromatic (PAN) satellite images.
Features¶
- 5 Classic Methods: Brovey, IHS, SFIM, Gram-Schmidt, HPF
- 7 Deep Learning Models: From simple CNNs to Transformers
- Advanced Loss Functions: L1, MSE, SSIM, SAM, Perceptual
- Attention Mechanisms: CBAM, SE blocks, Cross-attention
- Multi-scale Architectures: Feature pyramid networks
- Transformer Models: PanFormer with window attention
- Quality Metrics: PSNR, SSIM, SAM, ERGAS
- GeoTIFF Support: Preserves geospatial metadata
Quick Example¶
from models import create_model, create_loss
import torch
# Create model
model = create_model('panformer_lite', ms_bands=4)
# Create loss function
criterion = create_loss('spectral_focus')
# Run inference
ms = torch.randn(1, 4, 256, 256)
pan = torch.randn(1, 1, 256, 256)
fused = model(ms, pan)
Benchmark Results¶
Results on test dataset (100 epochs):
| Model | PSNR (dB) | SSIM | SAM | ERGAS |
|---|---|---|---|---|
| SFIM (classic) | 30.30 | 0.828 | 0.02 | 5.50 |
| PanNet | 30.79 | 0.839 | 0.04 | 2.41 |
| PanNetCBAM | 30.35 | 0.828 | 2.13 | 5.47 |
| PanFormerLite | 34.62 | 0.908 | 8.48 | 3.37 |
License¶
This project is licensed under the MIT License.