54
ERDC TR-05-1
Non-vegetated Pixels
Vegetated Pixels
Figure 32. Primary image segments after NDVI masking.
5.1.2.
Unsupervised Clustering
ERDAS Imagine image processing software contains a standard nonhierar-
chical clustering routine. This unsupervised classification algorithm was applied
to both the non-vegetated and the vegetated four-band, multispectral image seg-
ments created by the NDVI masking procedure. Imagine allows the analyst to
determine the total number of land cover classes (or clusters) classified within the
input image. For this study, eight thematic classes were delineated within each of
the primary segments (vegetated and non-vegetated). The selection of eight
classes for each of the two image segments was based on field experience at sites
similar to the reference wetlands. A subjective evaluation of the general land
cover structure and relative abundances of the vegetated and non-vegetated sur-
face features suggested that eight clusters would sufficiently classify each seg-
ment. Using the original four-band mosaic as an interpretative replacement for
ground-truth data, the analyst assigned a land cover designation to each of the
eight unique spectral classes was (Fig. 33). Finally, combining the segmented
thematic images produced a complete land cover classification of the area of in-
terest (Fig. 34).