PCA Image Fusion
Transforms multispectral bands into principal components, injects high-resolution detail into the dominant component, then reconstructs spectral bands.
C vk=λkvk
Best used when: a deterministic component-based fusion is preferred.
Image Fusion
Fuse low-resolution multispectral information with a high-resolution raster using component-based or learned self-supervised fusion methods.
Remote Sensing CoreEnd-User DocumentationTheory + Formula + Parameters1. What This Feature Does
Fuse low-resolution multispectral information with a high-resolution raster using component-based or learned self-supervised fusion methods.
Cloud workflow: choose data → configure the scientific method → run → review the resulting layer/report. No programming is required.
2. Recommended Workflow
Choose multispectral and high-resolution inputs.
↓
Choose bands and resampling.
↓
Select PCA, Neural Network, or KAN fusion.
↓
Configure training/sample controls for learned methods.
↓
Review spatial detail and spectral consistency.
3. Theory, Methods & Equations
Neural Network Autoencoder Fusion
Learns a nonlinear representation/reconstruction from degraded training pairs so high-resolution spatial information can be fused with multispectral content.
z=fenc(x), ŷ=fdec(z)
L=(1/n)Σ||y−ŷ||²
Best used when: scene-specific nonlinear fusion is useful and enough pixels are available for self-supervised training.
KAN Autoencoder Fusion
Uses KAN-style learnable univariate functions instead of conventional fixed activation edges for nonlinear fusion.
f(x)≈ΣqΦq(Σpφqp(xp))
Best used when: advanced learned image fusion is being explored.
4. Input Data
| Input | Type | Requirement | Description |
|---|---|---|---|
| Input Rasters | raster_stack | Required for selected method | Add one or more raster layers. Select the bands to include from each layer. All selected bands are aligned and used as the input feature dimensions. |
5. Parameters Available in the Application
| Parameter | Default | Choices / Range | How to Use It |
|---|---|---|---|
| Fusion Algorithm | — | PCA Image Fusion, Neural Network Autoencoder, KAN Autoencoder | Select the dimensional-fusion algorithm first. Only parameters for that algorithm are shown. |
| Output Components | 1 | 1 to 64 | Number of fused output bands/components. Default 1 combines all selected input dimensions into one output component. Shown for: pca, neural_network, kan |
| Alignment Resampling | bilinear | Nearest Neighbour, Bilinear, Cubic, Lanczos | Rasters after the first input are aligned to the first raster grid. Shown for: pca, neural_network, kan |
| Standardize Bands | Off | — | False uses covariance-style PCA on original band scales. Enable to perform correlation-style PCA after standardization. Shown for: pca |
| PCA Fitting Samples | 200000 | 1000 to 5000000 | Shown for: pca |
| Maximum Training Samples | 60000 | 1000 to 1000000 | Shown for: neural_network, kan |
| Prediction Batch Size | 200000 | 1000 to 2000000 | Shown for: neural_network, kan |
| Neural Network Encoder Layers | [64, 32] | — | Encoder hidden layers before the latent component layer. The decoder is mirrored automatically. Shown for: neural_network |
| Neural Network Activation | relu | relu, tanh, logistic | Shown for: neural_network |
| Neural Network Max Iterations | 300 | 20 to 5000 | Shown for: neural_network |
| Neural Network Learning Rate | 0.001 | 1e-06 to 1.0 | Shown for: neural_network |
| Neural Network Training Batch Size | 512 | 8 to 65536 | Shown for: neural_network |
| KAN Encoder Layers | [16, 8] | — | KAN encoder widths before the latent component layer. The decoder is mirrored automatically. Shown for: kan |
| KAN Grid Size | 5 | 2 to 32 | Shown for: kan |
| KAN Spline Order | 3 | 1 to 5 | Shown for: kan |
| KAN Epochs | 150 | 10 to 3000 | Shown for: kan |
| KAN Learning Rate | 0.01 | 1e-06 to 1.0 | Shown for: kan |
| KAN Training Batch Size | 1024 | 8 to 65536 | Shown for: kan |
| KAN Weight Decay | 0.0 | 0.0 to 1.0 | Shown for: kan |
| KAN Device | auto | auto, cpu, cuda | Shown for: kan |
| Random State | 42 | — | Shown for: pca, neural_network, kan |
6. Output & Interpretation
The result should be interpreted according to the selected method and the physical meaning of the input data. Preserve source units, coordinate reference information, NoData meaning, acquisition date, and preprocessing level when comparing results.
Scientific interpretation: an algorithm can produce a numerically valid result even when the input data are unsuitable. Always check masks, units, sensor characteristics, spatial resolution and reference data.
7. Best Practices
- Use analysis-ready inputs and remove invalid/cloud/noise artifacts that are not part of the target phenomenon.
- Choose parameters from the physical scale of the data, not only from visual appearance.
- Keep categorical and continuous rasters conceptually separate when selecting interpolation/resampling methods.
- Compare the result with the source image and independent reference information.
- Document the settings used when results will be compared across dates, sensors or study areas.