Home > Publications > Doctoral dissertation & Master thesis
Domestic Conference
Title Simple Fusion with Low Complexity for Distributed Multitarget Classification in Wireless Sensor Networks
Degree MS
Author Kwanglak Jung
Advisor Kiseon Kim
Graduation Date 2007.08.24 File
    Date 2013-09-11 20:36
Wireless sensor networks are one of the core technologies in the establishment of
ubiquitous networks. WSNs have applications in a wide variety of ?elds such as the
military, environmental monitoring, gathering sensing information in inhospitable lo-
cations, the observation or tracking of objects, and dangerous works. In practical
applications, it is important to grasp which targets are placed on a desired ?eld in
WSNs. Moreover, WSNs have a limited sensing range and power compared with other
systems. Thus, target classi?cation methods that minimizes power consumption are
required. There is ongoing research into classifying single targets, but in fact there can
be many targets and sensors distributed in sensing ?elds in WSNs. For this reason,
multiple target classi?cation also needs be considered.
The classi?cation scheme for multiple targets is divided into two types: centralized
and distributed classi?cation. While feature vectors are directly fused in centralized
classi?cation, hard decisions passed by local classi?ers are ?nally fused in distributed
classi?cation. Generally, decision fusion is used in distributed classi?cation. Since de-
cision fusion jointly processes combining low dimensional (i.e. scalar decisions) data
from decisions combined in node measurements, decision fusion has lower communica-
tion burden and training data requirement. Due to lower power consumption as a result
of this low communication burden in the transmission, we commonly use distributed
classi?cation as the preferred scheme with multiple targets.
However, distributed classi?cation with multiple targets has one critical problem.
광주과학기술원 한·러 MT-IT 융합기술연구센터 광주과학기술원정보통신공학부