17-03-2014, 10:46 AM
Abstract— Landmines are significant barrier to financial,
economic & social development in various parts of the world.
The demand of dependable, trustworthy, intelligent diagnostic
systems in the field of landmines detection has been increasing
rapidly. Metal detectors used in mine decontamination, cannot
differentiate a mine from metallic debris where the soil
contains large quantities of metal scrap & cartridge cases, so a
device is required that will reliably confirm that the ground
being tested does not contain an explosive device, with almost
perfect reliability. Human experts are unable to give belief &
plausibility to the rules devised from the huge databases. In
this paper two combined classifiers have been discussed. In the
first classifier Hebb Net learning is used with rough set theory
and in the second one Fuzzy filter neural network is used with
the rough set theory. Rough sets have been applied to classify
the landmine data because in this theory no prior knowledge of
rules are needed, these rules are automatically discovered from
the database. The rough logic classifier uses lower & upper
approximations for determining the class of the objects. The
neural network is for training the data, and has been used
especially to avoid the boundary rules given by the rough sets
that do not classify the data with cent percentage probability.
I. INTRODUCTION
Modern mines have minimal metal content to make
them harder to detect. A considerable challenge exists with
the processing of noisy signals and images sent by the
sensory system. Useful information can only be derived
with a carefully designed sensor fusion system backed by
strong signal and image processing algorithms. The control
algorithms must make sure the mine detector is able to
manipulate in unknown, uncertain environments, it should
be stable, and it should be able to learn from interaction and
can improve performance over time. Complex application
problems, such as reliable monitoring and diagnosis of
industrial plants, are likely to present large numbers of
features, many of which will be redundant for the task at
hand [1,2] . The landmine databases are large, and tracing
general knowledge from databases is known to be the most
difficult part of creating a knowledge-based system. The
most common approach to developing expressive and
human readable representations of knowledge is the use of
if–then production rules [3]. Yet, real life problem domains
usually lack generic and systematic expert rules for mapping
feature patterns onto their underlying classes.
The paper aims to induce low-dimensionality rule sets
from historical descriptions of domain features which are
often of high dimensionality. A common disadvantage of
techniques applied is their sensitivity to high
dimensionality, i.e., Principle Component Analysis
irreversibly destroys the underlying semantics of the feature
set [4]. Most semantics-preserving dimensionality reduction
(or feature selection) approaches tend to be domain specific,
i.e., relying on the use of well-known features of the
particular application domains.
Given a dataset with some feature values, it is possible
to have a subset of the original features using Rough Set
Theory [5,6] that are the most informative; all other features
can be removed from the dataset with minimal information
loss. Rough Set Theory is an alternative approach that
preserves the underlying semantics of the data while
allowing reasonable generality. It is, therefore, desirable to
develop this technique to provide the means of data
reduction for crisp and real-valued datasets which utilizes
the extent to which values are similar. And the neural
networks have been used in the training of data and the
classification of objects which came under the boundary
rules of the rough sets. In this paper two combined
classifiers have been discussed. In the first classifier Hebb
Net learning is used with rough set theory and in the second
one Fuzzy filter neural network is used with the rough set
theory.
economic & social development in various parts of the world.
The demand of dependable, trustworthy, intelligent diagnostic
systems in the field of landmines detection has been increasing
rapidly. Metal detectors used in mine decontamination, cannot
differentiate a mine from metallic debris where the soil
contains large quantities of metal scrap & cartridge cases, so a
device is required that will reliably confirm that the ground
being tested does not contain an explosive device, with almost
perfect reliability. Human experts are unable to give belief &
plausibility to the rules devised from the huge databases. In
this paper two combined classifiers have been discussed. In the
first classifier Hebb Net learning is used with rough set theory
and in the second one Fuzzy filter neural network is used with
the rough set theory. Rough sets have been applied to classify
the landmine data because in this theory no prior knowledge of
rules are needed, these rules are automatically discovered from
the database. The rough logic classifier uses lower & upper
approximations for determining the class of the objects. The
neural network is for training the data, and has been used
especially to avoid the boundary rules given by the rough sets
that do not classify the data with cent percentage probability.
I. INTRODUCTION
Modern mines have minimal metal content to make
them harder to detect. A considerable challenge exists with
the processing of noisy signals and images sent by the
sensory system. Useful information can only be derived
with a carefully designed sensor fusion system backed by
strong signal and image processing algorithms. The control
algorithms must make sure the mine detector is able to
manipulate in unknown, uncertain environments, it should
be stable, and it should be able to learn from interaction and
can improve performance over time. Complex application
problems, such as reliable monitoring and diagnosis of
industrial plants, are likely to present large numbers of
features, many of which will be redundant for the task at
hand [1,2] . The landmine databases are large, and tracing
general knowledge from databases is known to be the most
difficult part of creating a knowledge-based system. The
most common approach to developing expressive and
human readable representations of knowledge is the use of
if–then production rules [3]. Yet, real life problem domains
usually lack generic and systematic expert rules for mapping
feature patterns onto their underlying classes.
The paper aims to induce low-dimensionality rule sets
from historical descriptions of domain features which are
often of high dimensionality. A common disadvantage of
techniques applied is their sensitivity to high
dimensionality, i.e., Principle Component Analysis
irreversibly destroys the underlying semantics of the feature
set [4]. Most semantics-preserving dimensionality reduction
(or feature selection) approaches tend to be domain specific,
i.e., relying on the use of well-known features of the
particular application domains.
Given a dataset with some feature values, it is possible
to have a subset of the original features using Rough Set
Theory [5,6] that are the most informative; all other features
can be removed from the dataset with minimal information
loss. Rough Set Theory is an alternative approach that
preserves the underlying semantics of the data while
allowing reasonable generality. It is, therefore, desirable to
develop this technique to provide the means of data
reduction for crisp and real-valued datasets which utilizes
the extent to which values are similar. And the neural
networks have been used in the training of data and the
classification of objects which came under the boundary
rules of the rough sets. In this paper two combined
classifiers have been discussed. In the first classifier Hebb
Net learning is used with rough set theory and in the second
one Fuzzy filter neural network is used with the rough set
theory.