Using Non-authoritative Sources During Emergencies in Urban Areas - Computational Approaches for Urban Environments - page 338

0 1

X

n

@ A

o D K w j x j (14.1)

j D1

where K is a constant, is a nonlinear function, w j are the weights assigned by the
network, and is a threshold. The network takes inputs x and produces a response
o i from the output units i . The outputs are either o i D 1 if the neuron i is active
for the input x or o i D 0 if it is inactive. The network learns the weights through
iterative training and will converge when there is no change from one iteration to
the next.
The trained network can then be used for the classification of a new dataset. A
feedforward artificial neural network was implemented for this work using the R
statistical package (Venables and Ripley 2002 ).
The goal is to classify each pixel as being flooded or not flooded. The neural
network classifier is trained using the data layers from October 29 and tested on
the October 30 layers. Figure 14.6 illustrates the training of the neural network
classifier (blue lines) using available sources of remote sensing and authoritative
and non-authoritative data from October 29. The data are first preprocessed (e.g.,
georeferenced and interpolated) to create individual flood extent estimations which
are fed into the neural network to create a classifier. The operational step uses this
classifier along with data collected from the subsequent day, October 30. These data
(green lines) are preprocessed and then passed through the trained classifier to create
a flood extent map.

Fig. 14.6 Illustration of the application of a neural network classifier. The classifier is created
from training data ( blue lines ) and is then used to create a flood extent map by passing data from a
subsequent day ( green lines ) through the classifier