Image Processing Reference
In-Depth Information
restricted its utility to urban analysis, but for more than a decade MSS was the only
multi-spectral satellite sensor system available. Still, today it is used in historic
applications. The French SPOT (Systeme Probatoire d'Observation de la Terre) sys-
tem added much needed spatial detail for urban use. The three multispectral bands
(SPOT-X) roughly equivalent to TM bands 2, 3, and 4, have a nominal 20 m IFOV.
A very useful co-registered panchromatic (SPOT-P) band was set at 10 m. These data
have frequently been merged with multispectral data such as TM for urban studies
to gain spatial detail along with environmental (land cover) information. Landsat 7
was launched in 1999, with an improved sensor ETM+, with a co-registered
panchromatic 15 m band and improved thermal band at 60 m spatial resolution.
A new generation of satellite-based sensors has recently brought multispectral
remote sensing to 4 m (IKONOS) and panchromatic remote sensing to 1 m or less
(IKONOS, Quickbird, GeoEye, and others). These advances have added a robust
capability to V-I-S performance for large scale objectives. Meanwhile airborne
sensors have provided hyperspectral data: for example,
AVIRIS (Advanced Visible InfRared Imaging Spectro-
meter) yielding 224 spectral bands from the VIS and NIR/
SWIR range at a nominal 10 nm band width. Most of the
224 bands are of little value for urban analysis. However,
for certain regions of the spectrum, the narrow bands are
very useful, for example, in determining vegetation types
and conditions. The narrow bands also provide subtle
details of absorption features specific to certain minerals
in soil and help to distinguish concrete from asphalt.
Clearly, refined spectral resolution is vital to detection of the many cover types
in urban areas. Refined spatial resolution is also crucial, depending on the study
objective. Just as AVIRIS may provide more bands than needed, the new generation
of sensors may provide more spatial detail than needed, depending on study objec-
tives. Figure 4.7 shows that for a given test, accuracies of classification improved
from IKONOS (4 bands at 4 m), to TM (6 bands at 30 m), to AVIRIS (14 selected
bands at 10 m). The authors of Chapters 4 and 5 make the point that there are times
when spectral sensitivity outranks spatial resolution in achieving suitable classifica-
tion results.
In the following sections, a number of studies in V-I-S are summarized to illustrate
the utility of different datasets for various applications with various sensors of differ-
ent spectral and spatial resolutions. Also illustrated are various analytical techniques.
refined spatial
and spectral
resolution of new
remote sensing
systems add
robust capability
to the V-I-S
model
performance
6.4
Morphological Application of the V-I-S Model
This section applies the model to the morphology of cities. Five examples are given
with increasing spectral resolution (Table
6.3
). Spatial resolution varies from 30 m
to 10 m. Four examples employ sub-pixel analysis and one example employs artificial
neural network methods.
