# Linear Unmixing

Linear unmixing models every pixel's spectrum as a weighted sum of a few
**endmember** spectra (the pure materials in the scene) and solves for the
weight — the *abundance* — of each endmember in that pixel.

## How it works

For each pixel the tool solves a least-squares fit (the normal equations) of
the pixel spectrum against the stack of endmember spectra. The output is a new
dataset with one **abundance band per endmember** (in the order you listed
them) plus a final **RMSE band** giving the per-pixel reconstruction error, so
you can see where the chosen endmembers fail to explain the data.

Bands flagged bad in *either* the dataset or an endmember are excluded from the
fit. Endmembers must share the input's wavelength grid.

**Sum to Unity** (optional) softly forces each pixel's abundances to add up to
1, which is appropriate when every material is assumed present. The spin box
sets the penalty weight — larger values enforce the constraint more strongly.

The result is added as a new dataset named `Linear Unmix: <source>, N endmembers`.

## Using the tool

1. Choose an **Input Dataset**.
2. Build the **Endmembers** list (at least two): **Add Collected Spectrum** to
   pick from spectra collected in-app, or **Import Spectrum** to load them from
   a text file. Use the trash button on a row to remove it.
3. Optionally tick **Sum to Unity** and set its weight.
4. Click **OK**. The run proceeds in the background.

Click **View Past Runs** to revisit a previous run — it re-plots that run's
endmembers and resurfaces its input and output datasets.