Starting the n-D Visualizer with a pre-clustered result gives you a starting point for interactively rotating and refining the clusters of pixels into class groups. To use this feature, you must have MNF and PPI files for your data.

The pre-clustered result is a “first cut” selection of potential endmembers, shown as colored points in the data cloud. The pre-clustering operation attempts to find the corners of the data cloud in n-D space, based on the shape of the scatter plot and the related PPI scores of the pixels. Since the purest pixels are at the convex corners of the data cloud, based on the linear mixed pixel model, these are the pixels you want to identify and group together.

The algorithm first finds the pixel with the highest PPI score, using it as the seed for the first cluster. Then it iteratively creates more clusters until either the number of clusters is one more than the number of MNF bands used, or the inherent dimensionality of the data is reached. A second phase of clustering groups pixels that are near these corner points, reconciles clusters that may need to be joined, and checks the dimensionality of the simplex formed by the cluster means.

While pre-clustering is a useful tool, it cannot outperform a skilled human willing to take the time to understand the data, interactively explore it, and define corner clusters manually. As such, you should look at the results of the pre-clustering as a starting point for interactive validation, editing, and modification. Rotate the data cloud in the n-D Visualizer and modify the endmembers, as needed, using the n-D Visualizer controls.

  1. From the Toolbox, select Spectral > n-Dimensional Visualizer > n-Dimensional Visualizer Auto Cluster. The n-D Precluster Input MNF File dialog appears.
  2. Select the MNF input file. Click OK. The n-D Precluster Input PPI Band dialog appears.
  3. Select the PPI input file. Click OK. The n-D Precluster Parameters dialog appears.
  4. Enter a Maximum Number of Input Pixels value to use in the n-D Visualizer.
  5. Smaller numbers animate faster in the n-D Visualizer and show only the purest pixels; larger numbers give a better overall picture of the scatter plot, but they animate more slowly and may make selection of the corners more difficult. A threshold is automatically applied to the PPI image to obtain the best PPI pixels to use in the n-D Visualizer without exceeding the selected maximum.

    The n-D Visualizer and n-D Controls dialogs appear. The precluster results are shown as colored pixels in the n-D Visualizer.

  6. Rotate the data cloud to assess the results and modify them as needed.