Environmental Engineering Reference
In-Depth Information
(a)
(b)
(c)
(d)
FIGURE 3.5
Results for part of the Bam test site: (a) the initial classification maps from the pre-event Quickbird image; (b)
the final classification map from the pre-event scene; (c) the final classification map from the post-event scene; (d) visual
representation of the thresholded BDI values, compared with the validation data that are in blue and red dots (see the relevant
text for the explanation).
data has been made available as a vector file containing the center
location of each building, with an attribute defining its damage
level according to the five grades of the European Macroseismic
Scale (EMS), from ''undamaged'' to ''heavily damaged''. For this
work, building damage classes were reduced to ''strongly dam-
aged'' (red dots, EMS classes 4 and 5) and ''lightly damaged''
(blue dots, EMS class 1 to 3). The confusion matrix between the
reference and the BDI-derived damage assessment in Table 3.2
shows a really good accuracy (nearly 80%), with greater preci-
sion, as expected, for strongly damaged buildings. The BDI seems
then to be a good measure of damage extent, even in presence
of four main limiting factors that do not directly depend on the
index itself:
some buildings could not be selected because of a lack of
precision in classification (there are also multiple objects
classified as a unique building);
●
performances are limited by differences in atmospheric and
acquisition conditions, since they increase the number of false
alarms;
●
the confusion matrix takes into account just true negatives
and not false positives;
●
some objects are considered errors because of little co-
registration misplacements between the output map and the
ground truth.
●
Search WWH ::
Custom Search