nNMF Explorer
Non-negative Matrix Factorization (nNMF) decomposes a hyperspectral datacube into spatial component maps and corresponding spectral signatures.
IDCubePro 2026 - nNMF Explorer
Overview
Non-negative Matrix Factorization (nNMF) decomposes a hyperspectral datacube into spatial component maps and corresponding spectral signatures.
The decomposition is represented as:
W = spatial component contributions The original data are approximated by W × H.
Select the number of nNMF components before running the analysis.
A small number produces a simpler representation of the dataset, while a larger number can capture more spectral variation.
Click Run nNMF to perform the decomposition.
Negative input values are clipped to zero before analysis because nNMF requires non-negative data.
After processing, the component maps, spectral signatures, reconstruction, and RMSE become available.
RMSE measures the difference between the original hyperspectral data and the nNMF reconstruction.
Lower RMSE indicates a reconstruction that more closely reproduces the original dataset.
The Original panel displays the currently selected spectral band from the original hyperspectral cube.
Move the wavelength slider to inspect different spectral bands.
When wavelength metadata are available, the current wavelength is displayed together with the band number.
The component panel displays the spatial W map associated with each nNMF component.
Click a component map to select that component.
You can also select a component from the Component dropdown.
The Component Spectrum panel displays the H spectral signature for the currently selected nNMF component.
The horizontal axis represents wavelength when available.
Displays the reconstruction produced by the selected component alone.
Displays the reconstruction generated using all nNMF components.
Displays the reconstruction using components 1 through the currently selected component.
Displays the difference between the original datacube and the full nNMF reconstruction.
Move the mouse over the Original image after nNMF has been calculated.
The Pixel Spectra panel compares:
This provides a local assessment of reconstruction quality at individual image pixels.
Exports the complete reconstructed hyperspectral datacube generated from W × H.
The output retains the original spectral wavelength vector.
Exports the contribution from the currently selected nNMF component as a hyperspectral datacube.
This represents W(:,k) × H(k,:) reconstructed back into the original spatial and spectral dimensions.
The Quality Curve evaluates reconstruction error as the number of components increases.
For speed, the calculation uses a random sample of image pixels when the dataset is large.
The graph displays RMSE versus number of components.
A flattening of the curve can help identify a reasonable number of components for the dataset.
Exports the complete nNMF decomposition.
W - spatial component coefficients H - spectral component signatures Wmaps - W reshaped into spatial component images Hspectra - spectral component signatures Wavelengths - spectral wavelength vector Image_approx - reconstructed hyperspectral cube
1. Select an initial number of components.
3. Inspect the spatial component maps.
4. Click individual components and examine their spectral signatures.
5. Compare Selected Component, Full Reconstruction, Cumulative Components, and Residual views.
6. Hover over the Original image to compare pixel spectra.
7. Use Quality Curve if you want to evaluate the number of components.
8. Save a reconstruction or individual component if needed.
9. Export W / H for quantitative or downstream analysis.