Use Spectral Feature Fitting (SFF) to compare the fit of image spectra to reference spectra using a least-squares technique. SFF is an absorption-feature-based methodology. The reference spectra are scaled to match the image spectra after the continuum is removed from both datasets.
For more information, see Spectral Tools References and Spectral Feature Fitting Results.
- From the Toolbox, select Spectral > Mapping Methods > Spectral Feature Fitting. The Spectral Feature Fitting Input File dialog appears.
- Select the input file and perform optional spatial subsetting, and/or masking, then click OK.
Input spectra for SFF should be from a continuum-removed perspective; there must be some degree of concavity within the spectral “curve” to avoid generating an error.
If the selected input file is not continuum-removed, ENVI removes the continuum on-the-fly. However, the function executes much slower.
- In the Input File dialog, select Spectral Subset. The File Spectral Subset dialog appears.
- Select bands to subset around the region containing the absorption features of interest.
- Click OK.
- Click OK in the Input File dialog. The Endmember Collection:Feature Fitting dialog appears.
- Import the reference spectra. See Import Spectra for details.
- Click Apply. The Spectral Feature Fitting Parameters dialog appears.
- Use the toggle button to switch between Output separate Scale and RMS Images or Output Combined (Scale/RMS) Image. For more information, see Working with SFF Results.
- Select output to File or Memory.
- Click OK. ENVI adds the resulting output to the Layer Manager.
Spectral Feature Fitting Results
A scale image and RMS image or a combined “fit” (scale/RMS) image is output for each reference spectrum. The image is a measure of absorption feature depth, which is related to material abundance. The brighter pixels in the scale image indicate a better match to the reference material in those pixels (for areas with a low RMS error). However, a large scale value (> 1) can result if incorrect reference endmembers are input or if the incorrect wavelength range is used. The image and reference spectra are compared at each wavelength in a least-squares sense, and the RMS error is calculated for each reference spectrum. Dark pixels in the RMS error image indicate a low error. You can use the RMS errors and scale image results to locate areas that best match the reference spectrum.
Working with SFF Results
To view areas in the image that best match the reference spectrum, use 2D scatter plots to plot the scale versus RMS.
Another way to produce results that show the distribution of the reference material is to use the “fit” (scale/RMS) image. Higher fit values indicate better matches to the reference spectrum.
To generate an output classification-like image, select Classification > Post Classification > Rule Classifier from the Toolbox to threshold the output images using a maximum threshold value.