Image Processing Reference
In-Depth Information
Fig. 4.8 Overall accuracies and KAPPA coefficient for different sensor configurations and varying
spatial resolution
resolution the classification performance of IKONOS/LIDAR drops below the
AVIRIS accuracy. Hence, AVIRIS data analyses are less sensitive to changes in
spatial resolution. Although the trends certainly vary for different land cover
classes, IKONOS and LIDAR data strongly depend on the accurate representation
of individual urban land cover objects and should only be used in fine spatial reso-
lution. If only coarse spatial resolution data are available fine spectral resolutions
should be preferred for urban land cover mapping.
It should be noted that these results reflect a purely pixel-based spectral mapping
perspective. Thus, if the remote sensing mapping objective is focused on the spatial
and geometric properties of land cover structures (e.g., the shape and size of buildings),
fine spatial resolution data on the order of 3-5 m are required for a clear representa-
tion of the urban environment (Jensen and Cowen 1999 ). Also, the use of object-
oriented, segmentation image analysis approaches can add an additional level of
information to the image classification and help to resolve some of the limitations
shown here for spatial-spectral resolution dependent mapping approaches.
Chapter Summary
This chapter has provided an overview of the spectral character of urban
attributes. Urban areas with roofing materials, pavement types, soil and
water surfaces, and vegetated areas represent a large variety of surface
compositions. The spectra of urban spectral libraries reflect these properties
in characteristic spectral signatures and related absorption features. Specific
urban land cover types show low spectral separability over the whole spectral
range from 350 to 2,400 nm (i.e., specific roof and road types). This results
in lower remote sensing mapping accuracies. The most suitable wavelengths
for separation of urban land cover emphasized specific spectral features that
provide the best separation and highlight the importance of hyperspectral
urban remote sensing. Spectral limitations exist for current multispectral
sensors where the location and broadband character of the spectral bands
Search WWH ::




Custom Search