Spectral Feature Fitting (SFF)

Spectral Feature Fitting (SFF)#

SFF measures how well each pixel matches a reference spectrum by comparing their absorption features — the dips a material produces at characteristic wavelengths — rather than the overall spectrum.

How it works#

SFF isolates absorption features with continuum removal: it divides out the smooth “continuum” (the broad upper envelope of the spectrum), flattening the overall slope and brightness so only the absorption dips remain. The result is inverted (1 − continuum-removed) so each absorption band becomes a positive peak. This is done for both the pixel spectrum and the reference, over the chosen wavelength range.

SFF then fits the reference’s features to the pixel’s by finding a single scale factor (via least squares) that best matches their depths. Two numbers come out per pixel:

  • RMS error — the residual after fitting; how well the feature shapes agree. Lower RMSE means a better match, and an RMSE below the threshold marks a detection.

  • Scale — the fitted depth multiplier, a rough indicator of how strongly the feature is expressed (loosely, abundance).

Because it keys on absorption shape, SFF works best when you restrict the wavelength range to a known diagnostic feature. References are interpolated onto the target’s wavelength grid, bad bands are dropped.

In Image Cube mode each run produces three datasets, one band per reference spectrum:

  • SFF CLS, Img: <source> — boolean classification (RMSE < threshold).

  • SFF RMSE, Img: <source> — the fit error at every pixel.

  • SFF SCALE, Img: <source> — the fitted scale (feature depth).

Using the tool#

Inputs work the same way as Spectral Angle Mapper:

  1. Target — choose Image Cube (or Spectrum) and select the dataset (or spectrum) to analyze.

  2. Wavelength range — set the min/max and units; for SFF this should bracket the absorption feature of interest.

  3. References — check one or more reference spectra. Add them via Add Library, Add Spectrum (from a text file), or Add Collected Spectrum; a default USGS mineral library is preloaded. Each reference has its own threshold (Initial RMSE, default 0.03), seeded from the method threshold but individually overridable.

  4. Click Run SFF. Image-cube runs proceed in the background; spectrum runs open a ranked details table with a comparison plot.