SEARCH
0-9 A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Prev | Current Page 260 | Next

Yingshu Li, My T. Thai, and Weili Wu

"Wireless Sensor Networks and Applications"


Let sensor S0 be the sensor performing classifier-based edge detection and
PAs be the set of sensors in its probing area. A line with the above characteristic
must have the highest classifier score Js0 :
155
Ren-Shiou Liu, Lifeng Sang, and Prasun Sinha
Fig. 4. A simple linear classifier cuts the probing area S0 into two regions such that
sensors with di?®erent values of event predicates are at di?®erent sides.
Js0 (a, b, c) = ????????????
X 8s2PAs0
VsSN(axs + bys + c)????????????
(16)
where
SN(x) = ??±??????
??’1 if x < 0,
0 if x = 0,
1 if x > 0.
(17)
The proof is straightforward. Based on (16) and (17), for all s above
the line, VsSN(axs + bys + c) > 0. If most of the sensors with 1 values of
event predicates lie on the same side, say above the line, then Js will reach a
maximum value. However, a partition is regarded as valid if and only if the
optimal line Lopt(a, b, c) satisfies axs0+bys0+c p(a2+b2) ?· r, which means that the line
lies within a predefined tolerance radius.
Generally speaking, as the radius of probing area or sensor density increases,
the more accurately the event boundary can be detected. This applies
to the statistical approach, image processing approach and the classifier-based
approach. However, the classifier-based approach has a far more qualitative
characteristic than the others. When the probing area or sensor density increases,
the unwanted detection rates for the statistical and image processing
approaches also increase, while it??™s a decrease for the classifier-based approach,
which means a thinner edge can be obtained.


Pages:
248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272