Image Processing Reference
In-Depth Information
The separability scores emphasize the generally higher degree of separability for
green vegetation, gravel roofs and roads, and red tile roofs. Lower average and lower
minimum separability are obvious between asphalt roads,
parking lots, and specific types of roof materials such as com-
posite shingle, tar and gray tiles. These classes are spectrally
very similar on the material scale. The spectral similarity of
these classes has already been observed in their spectral sig-
natures (Fig. 4.2 ). Concrete roads have fairly high average
separability scores for all classes. However, their low mini-
mum separability with specific classes such as bare soil, park-
ing lots and several roof types emphasizes some degree of
spectral similarity caused by the large heterogeneity of concrete road surfaces.
Figure 4.4 also emphasizes that the spectral confusion of road materials is mostly
with non-road surface types, e.g., specific roofs, bare soil and parking lots. Wood
shingle roofs and NPV represent another pair of classes with good average but low
minimum separability. These materials are similar and some confusion between
these classes may be observed in remote sensing data (Fig. 4.5 , see also Fig. 4.3 ).
Basically, the analysis of spectral separability quantifies low spectral separation
between some urban land cover types. Particularly, large confusion exists between
specific roof types, and between some road materials, specific roof materials, bare
some urban
land cover types
can hardly be
separated
based on their
spectral
characteristics
Fig. 4.5 Most suitable spectral bands for urban mapping derived from the ground spectral library
and hyperspectral AVIRIS data compared to spectral signatures of several urban land cover types
and the spectral coverage of LANDSAT ETM satellite sensor ( gray in the background ). The major
water absorption bands (1,340-1,480 and 1,770-1,970 nm) are linearly interpolated
Search WWH ::




Custom Search