|ENVI User's Guide: Spectral Tools|
Use ENVI's Spectral AnalystTM to help identify materials based on their spectral characteristics. The Spectral Analyst uses ENVI techniques, such as binary encoding, Spectral Angle Mapper, and Spectral Feature Fitting to rank the match of an unknown spectrum to the materials in a spectral library (see Using Binary Encoding, Spectral Angle Mapper Classification, and Using Spectral Feature Fitting for method descriptions). You can also define your own spectral similarity techniques and add them to the Spectral Analyst (see "Spectral Analyst Functions" in the ENVI Programmer's Guide). The output of the Spectral Analyst is a list of the materials in the input spectral library ranked in order of best to worst match. An overall similarity score, and individual 0 to 1 scores for each method are reported. For more information, see Tips for Successful Use of the Spectral Analyst.
This function does not identify spectra, it just recommends likely candidates for identification. The results may change when the similarity methods used or weights are changed. It is still up to you to make the actual identification.
A plot window must be displayed.
The weighting factors can have any value. A perfect match to a library spectrum would have an output Score of the sums of the weights.
A recommended starting point is to use a weight of 1.0 in Spectral Feature FittingTM.
Use the Min and Max factors to indicate which values are considered a perfect match by scaling them to 0 to 1 (or 1 to 0) scores.
A SAM or SFF result less than or equal to the min value indicates a perfect match and gets a score of 1. A SAM or SFF result greater than or equal to the max value gets a score of 0.
For the binary encoding method, enter the min and max values as a percentage of bands correctly matched (0-1).
A binary encoding result less than or equal to the min value gets a score of 0 and a result greater than or equal to the max value a gets a score of 1.
The results of the similarity measures are listed. ENVI resamples the spectral library to match the spectral resolution of the input spectrum.
Use the Spectral Analyst window to open a new spectral library, edit the weights and min and max values, input x and y scale factors, and get input spectra from a Z-profile plot.
To use a specific wavelength range:
To display the input spectrum and a selected library spectrum (with the continuum removed) in a plot together, double click with the left mouse button on the library name in the Spectral Analyst list.
To open a new spectral library file to use in the comparisons:
To edit the methods' weighting factors and min and max values:
To enter or edit X and Y scale factors used to scale the input spectrum into the same space as the spectral library:
To enter spectra directly from a Z-profile window:
The Z Profile is updated and the spectral comparison information appears in the Spectral Analyst window. As you move the Zoom box in the Image window, the information in the Spectral Analyst changes accordingly.
If you have two images displayed, you must select Options Clear Auto Inputs before using the Spectral Analyst in the second display.
The output of the Spectral Analyst is a ranked or weighted score for each of the materials in the input spectral library. The highest score indicates the closest match and indicates higher confidence in the spectral similarity. Similar materials may have relatively high scores, but unrelated materials should have low scores.
The Spectral Analyst is based on use of spectral matching techniques that have specific requirements for successful operation. The items in this section summarize some of factors needed to effectively use the Spectral Analyst.
Many materials are similar in one wavelength range, yet very different in another range. For best results, use the wavelength range that contains the diagnostic absorption features. When a spectrum displays, the Spectral Analyst works over the range displayed in the plot being analyzed. If a sub-range is desired, use the middle mouse button in the plot to zoom in on the desired wavelength range before clicking Apply in the Spectral Analyst.
Determine whether materials have absorption features. If they do, Spectral Feature Fitting is probably the best method. Otherwise, Spectral Angle Mapper and/or Binary Encoding will give better results.
Materials of interest in the image spectra that are not in the reference spectral library will not be identified. If a material is not in the reference spectral library, materials similar to it may score relatively high.
High scores for multiple materials may indicate mixtures, particularly for minerals that have absorption features in different parts of the spectrum. Because of this, it is best to run the Spectral Analyst function on the results of an endmember determination- the "pure" endmember spectra.
Higher scores indicate higher confidence, because more of the rules were satisfied. Greater separation between adjacent scores indicates higher confidence in the similarity. For example a score of 0.98 for one material followed by a score of 0.96 for another material indicates that the unknown is very similar to both sets of rules, whereas a score of 0.98 for one material and a score of 0.50 for another material indicates a high probability that the spectrum represents the first material.
In many cases, the Spectral Analyst may list multiple identical scores for different materials in the rule base. This indicates that the Analyst can not discriminate the two materials under the conditions being used for identification. In that case, try a different wavelength range or use the different weighted methods to produce a unique answer.
In some cases, the Spectral Analyst may list a score of 0.000 (no match) for all materials in the library. This is a good indication that the material is not in the spectral library used, as well as not similar to other materials in the library.
Finally, examine the spectral ranking in the context of the image setting and known information. If a suggested identification seems invalid with respect to the known information, it is probably not the correct identification.
The Spectral Analyst tool is not foolproof. It is meant to be used as a starting point to put you on the right track towards identifying the materials in an image scene. Used properly and with a good spectral library, it can provide excellent suggestions for identification. Used blindly, it can produce totally erroneous results.
ENVI Online Help (August 12, 2005)