>  Docs Center  >  Using ENVI  >  Hyperspectral Material Identification

ENVI

Hyperspectral Material Identification

Hyperspectral Material Identification

Use the Hyperspectral Material Identification tool to attempt to identify unknown spectral signatures by comparing them to spectral libraries. This tool differs from the Material Identification tool in that it takes into account background statistics and endmembers from the image, and therefore can give a more accurate answer and spectra plots for rare or sub-pixel targets. Follow these steps to continue:

  1. Open and display an image to investigate.
  2. Select THOR > THOR Hyperspectral Material Identification from the Toolbox.

    The Hyperspectral Material Identification panel appears.

  3. Click Add Library and select an MRSL or SLI spectral library file to add known signatures that will be used to identify the input signatures.
  4. Click Select New Image to load the image into the Hyperspectral Material Identification tool. This will also compute the statistics and endmembers used for identification.
  5. In the Layer Manager, right-click the displayed layer, then click the Spectral Profile button to open the Spectral Profile window. Then, in the Image window, click on the pixel you want to identify, to plot it in the Spectral Profile.
  6. In the Hyperspectral Material Identification dialog, select the libraries to search in the Select Libraries to Search list, then click Select New Signature. A Material Identification dialog appears, with a list of all signatures currently open in plot windows.
  7. Select a spectrum from the list, then click OK. The Hyperspectral Material Identification tool will attempt to automatically determine the proper scale factor needed to scale the unknown signature into the same scale as the reflectance libraries. It will use the Adaptive Coherence Estimator (ACE) algorithm to compare the unknown signature to every signature in the designated spectral libraries. You can adjust the scale factor by entering a new value in the Input signature scale field and pressing the Enter key.

    The matching results are displayed at the bottom of the Hyperspectral Material Identification panel in the following columns:

    ACE: ACE measures if a spectrum is a good match for the spectrum in the selected image. Use ACE to determine candidate materials; that is, materials that may be contained in the pixel. ACE values range from -1 to 1 with scores close to 1 indicating a best match. The spectra in the table are initially sorted by ACE with the best matches listed on top. Results are color-coded by ACE as follows:

    Green: 0.9-1
    Yellow: 0.75-0.9
    Red: 0.5-0.75
    Gray: <0.5

    Likelihood: Likelihood measures how good a spectrum is in comparison to the other spectrum in the library. Use likelihood to distinguish between candidate materials. If the likelihood for spectra A is twice that of spectra B, then the probability of spectra A is twice that of the probability for B. The likelihoods will all sum to 1. Compare likelihood values between different materials to determine between those materials.

    Full Pixel Correlation: Full Pixel Correlation measures how good the pixel spectrum matches each library spectrum. Use this to determine candidate materials that are full pixel or “common” in the image. (Common materials include vegetation, soils, etc. that do not receive high ACE scores.) Correlation values range from -1 to 1 with scores close to 1 indicating a best match.

    Background-Removed Correlation: Background-Removed Correlation measures how good the background-removed pixel spectrum matches each library spectrum. Use this to determine candidate materials that have low abundance in the pixel. Correlation values range from -1 to 1 with scores close to 1 indicating a best match.

    Library Source: The library that contains the potential match for a candidate material.

  8. Click Export to ENVI Plot to export the plots to a Spectral Plot window.

    The plot area displays the unknown signature in red, the currently selected library signature in the results table in green, and the background removed correlation in blue. The background removed correlation signifies how good the background-removed pixel spectrum matches each library spectrum. The larger the band importance, the more effect that band has on enlarging the spectral angle between the unknown and library signature. You can turn off the background removed correlation by disabling the Plot background removed spectrum check box.

  9. At times, you may want the Hyperspectral Material Identification tool to ignore certain noisy or otherwise bad bands when comparing signatures. To designate which bands should be ignored, enable the Edit bad bands check box. The plot area will be shaded green for "good" bands and red for "bad" (ignored) bands. Click the red Bad Bands button, then click and drag in the plot area to designate those bands as "bad." Click the green Good Bands button, then click and drag in the plot area to designate those bands as "good." For an example of how to designate bad bands, see Material Identification.
  10. After specifying bad bands, disable the Edit bad Bands check box to compare signatures using the new set of good bands. Bad bands are blanked out in the plot area.
  11. If you are using an MRSL spectral library, you can view metadata for the selected signature. In the results table, select a signature, then click View Metadata. The Spectrum Metadata dialog appears.
  12. You can export the results table to an ASCII text file by clicking Export Table.



© 2019 Harris Geospatial Solutions, Inc. |  Legal
My Account    |    Store    |    Contact Us