Terrain Categorization

The results illustrated here were generated with a modified version of IMAGINE Subpixel Classifier. This software is not currently commercially available.

In response to a government request, IMAGINE Subpixel Classifier was enhanced to detect multiple materials simultaneously, unlike the commercial product that is used to search for one material at a time. This modified software can be used to categorize major terrain features for the purpose of landscape characterization or habitat analysis. It identifies the principal scene materials and the fraction of each material in each pixel.

Includes material © Space Imaging L.P.

This Quantitative Terrain Categorization is an unsupervised process. The computer identifies spectral groups and labels them as land or water. A human analyst can further label these output classes if desired. Computer analysis is done on a subpixel level, so more than one spectral class can be present in each pixel.

A separate image plane is generated for each of 16 spectral classes so that the amount of each material can be quantified as occupying a certain percent of the pixel. By combining these image planes, the researcher can see relationships between classes. There is significantly more information than in a traditional 16 class terrain categorization.

For example, one class may be identified as water while a second class characterized as trees. If both classes appear in the same pixel, as they may with this quantitative terrain categorizer, then an analyst may assess the area as forested wetland, an environmental unit not uniquely represented by the 16 spectral classes.

In the results illustrated here, the occurrences of six of the sixteen spectral classes are shown (classes 1, 2, 3, 10, 12 and 13). An analyst using aerial photography of a few sites has labeled class 12 as coniferous forest (green) and class 13 as deciduous forest (red). Those areas shown in orange contain both class 12 and 13, a mixture of both coniferous and deciduous trees. This mixture does not occur randomly throughout the scene, but rather indicates particular ecological units.