Multi-Target Classification Using Acoustic Signatures in Wireless Sensor Networks: A survey

Ahmad Aljaafreh & Ala Al-Fuqaha
Signal Processing-An International Journal (SPIJ), Volume (4): Issue (4) 175
Multi-Target Classification Using Acoustic Signatures in
Wireless Sensor Networks: A survey
Ahmad Aljaafreh aljaafreh@ieee.org
Electrical Engineering Department
Tafila Technical University
Tafila, 66110, P.O.Box 179, Jordan
Ala Al-Fuqaha ala.al-fuqaha@wmich.edu
Computer Science Department
Western Michigan University
Kalamazoo, MI 49008, USA
Abstract
Classification of ground vehicles based on acoustic signals using wireless sensor
networks is a crucial task in many applications such as battlefield surveillance,
border monitoring, and traffic control. Different signal processing algorithms and
techniques that are used in classification of ground moving vehicles in wireless
sensor networks are surveyed in this paper. Feature extraction techniques and
classifiers are discussed for single and multiple vehicles based on acoustic
signals. This paper divides the corresponding literature into three main areas:
feature extraction, classification techniques, and collaboration and information
fusion techniques. The open research issues in these areas are also pointed out
in this paper. This paper evaluates five different classifiers using two different
feature extraction methods. The first one is based on the spectrum analysis and
the other one is based on wavelet packet transform.
Keywords: Signal classification, feature extraction, distributed sensors, sensor fusion.
1. INTRODUCTION
Wireless sensor network (WSN) is a network of spatially distributed, densely deployed, and self
organized sensor nodes, where a sensor node is a platform with sensing, computation and
communication capabilities. WSN is an emerging technology because of the advances in
technologies of: Micro-Electro-Mechanical Systems (MEMS), Microprocessors, wireless
communication and power supply. New technologies provide cheap small accurate: sensors,
processors, wireless transceivers, and long-life batteries. Sensor node is the integration of all of
these technologies in a small board, like the ones in Fig. 3 part (b), it is called mote. Fig. 3 part (a)
shows the basic architecture of the mote. All of the above motivate researchers and practitioners
to design, deploy and implement networks of these sensor nodes in many applications. WSN has
the following characteristics: concern is about the data but not about the sensor node itself, low
cost, constrained power supply, static network, topology may change because of sensor node or
link failure, sensor nodes are prone to destruction and failure, dense deployment, self-
organization, and spatial distribution. WSN is used in many remote sensing and data aggregation
applications [1],[2]. Detection, classification, and tracking are the main signal processing functions

of the wireless sensor networks [3]. WSNs increase the covered area, redundancy of the sensors,
and decision makers, which improves the performance and reliability of the decision making. To
understand the work, design and operation of the WSNs see Refs. [4],[5]. Refs. [4],[6] categorizes
the applications and describes the implementation of the WSNs. A survey of the architecture and
sensor nodes deployment in WSNs is presented in Ref. [7]. WSN is a cost efficient technology.
However, it has some constraints. Limited energy, limited bandwidth, and limited computational
power are the main constraints of WSNs [8]. Therefore, to implement any digital signal
processing algorithm it needs to be an intelligent signal processing and decision making algorithm
with the following requirements: power efficiency, robustness, and scalability. In WSNs, observed
data could be processed at the sensor node itself, distributed over the network, or at the gateway
node. WSNs can be utilized for distributed digital signal processing [9]-[11]. Research in
classification in wireless sensor networks can be divided into two areas: hardware area
(platforms, sensors), and software area (signal processing algorithms, collaboration, and
networking techniques) [12]. The signal processing techniques and collaboration schemes that
are used in ground vehicle classification in WSN based on acoustic signals are surveyed, as in
Fig 2, in this paper. Target classification in WSN is to label or categorize a target that passing
through the area that is monitored by the WSN to one of a predefined classes based on an
extracted feature vector. Classification in WSNs can be considered as a process as in Fig. 4,
where a feature vector is extracted from the input signal, then classified, then the information is
fused to come up with the final decision. Most of the researcher are interested in improving the
performance of this process through selection and design an efficient tool, as in Table 1, for one
of the followings tasks :
• Feature Extraction
• Classification Techniques
• Information Fusion
The remainder of the paper is organized as follows. Section 2 presents the recent methods that
are used to extract features from the vehicle acoustic signals for single and multiple targets.
Section 3 discusses the classification techniques. Section 4 presents the information fusion
techniques. Section 5 outlines the the open research. And finally, conclusions are discussed in
section 6.
Reference Feature
Extractor
Classifier Classes
Number
Classification
Rate
Fusion
Method
[12] TESPAR ANN 2 up to 100% -
[13] and [14] DWT MPP 2 98,25% -
[15] HLA, PSD ANN 4 HLA: 92%,
PSD: 94%
-
[16] HLA ANN 18 88% running sum
[17] HLA MAP 6 89% -
[18] MFCC GMM, HMM
and ML
9 77%, 88% -
[19] FFT, DWT,
STFT, PCA
kNN, MPP 4 85%, 88% MRI
[20] STFT, PCA ANN 3 - -
[21] FFT, PSD,
AR
kNN, ML,
SVM
2 78% - 97% -
[22] DWT ANN 4 73% -
[23] CC HMM 9 96% -
[24] WPT LDA, CART 3 - -
[25] CWT ANN 6 95% -
[26] TVAR, PCA ANN 6 83%-95% -
[27] BHM CART 9 90% Decision
Fusion
[28] STFT, RID ANN, MVG 6 up to 87% -
[29] EE, PCA ANN, Fuzzy 5 up to 97% -

Logic
[30] AR ANN 4 up to 84% -
[31] FFT, PSD kNN - -% -
[32] FFT, WDT kNN 2 62% Dempsler-
Shafer, MV
[33] FFT Template
Matching
8 -% template
storing
[34] PSD kNN, ML 2 77%, 89% Distributed
Classification
[35] - kNN, ML,
SVM
2 69%, 68%,
69%
MAP
Bayesian,
Nearest
Neighbor,
Majority
Voting,
Distance-
based
[36] Harmonic
and
Frequency
Components
SVM 5 85% modified
Bayesian
(decision
level)
[37] Harmonic
set
MVG 3-5 70-80% -
[38] STFT,PCA C4.5, KNN,
PNN, SVM
4 60-93% -
[39] WPT CART - - -
[40] MFCC RNN 4 85% -
[41] PSD KNN, ML,
SVM
2 up to 97% -
[42] PSD, PCA SVM 3 up to 93% -
[43] MFCCs GMM 2 up to 94.1% CART
[44] FFT, WT KNN, MPP, K-
Means
3 95.5% -
[45] WPT ML, ANN 3 up to 98% -
[46] PSD ANN 4 up to 99% -
[47] WPT cascaded
fuzzy
classifier
(CFC)
3 - Dempster–
Shafer (DS)
Table 1: Recent feature extraction and classification techniques used for vehicle
classification based on acoustic signals.
2. FEATURE EXTRACTION OF ACOUSTIC SIGNATURE
Feature extraction is the most significant phase of the classification process. To classify an
object, a set of features of that object is extracted to label that object to one of a predefined
classes. This set of features is generated from a source signal as in Fig. 1. Feature extraction can
be considered as dimensionality reduction technique. In feature extraction certain transforms or
techniques are used to select and generate the features that represent the characteristic of the
source signal. This set of features is called a feature vector. Feature vectors could be generated
in time, frequency, or time frequency domain.

Figure 1: Classification block diagram.
Figure 2: Taxonomy of the techniques that are used in target classification using acoustic
signature in wireless sensor networks
.
Figure 3: Wireless sensor node examples in part (b) and the common architecture of a
senor node in part (a).

Figure 4: A summary diagram of the feature extraction, classification , and collaboration
algorithms that are used in vehicle classification using WSNs.
2.1 Time Domain
The computation of feature vector in time domain is usually simple. Ref. [48] discusses two time-
domain feature generation methods. The first method is based on the energy distribution of the
signal, where the energy of a short time window of the source signal is used to discriminate
between classes. The second method is based on counting the number of zero crossings of a
signal within a time interval. The energy envelope (EE) in time domain is considered in [29]. Time
Encoded Signal Processing and Recognition (TESPAR) is a method that is used in speech
waveform encoding. TESPAR is used in [12] to generate features from vehicle acoustic and
seismic signals. TESPAR is based on the duration and shape of the portion of the waveform that
is between two zero crossings.
Principal Component Analysis (PCA) is popular statistical tools that is used for dimensional
reduction. PCA is based on finding the principal eigenvectors of the covariance matrix of the set
of signals. PCA is used as a feature extraction method in [19, 20, 38, 42].
2.2 Frequency Domain
Frequency based feature generation methods, like Fast Fourier Transform (FFT), are common
approaches in vehicle classification [16], [20], [31], [33]-[35], [49]. In [31] Fast Fourier Transform
(FFT) and Power Spectral Density (PSD) are used to extract feature vectors. Similarly in [35], the
first 100 of 512 FFT coefficients are averaged by pairs to get a 50-dimensional FFT-based feature
vector with resolution of 19.375 Hz and information for frequencies up to 968.75 Hz. Ref. [34]
presents schemes to generate low dimension feature vectors based on PSD using an approach
that selects the most common frequency bands of PSD in all the training sets for each class. Ref.
[33] proposes an algorithm that uses the overall shape of the frequency spectrum to extract the
feature vector of each class. Principal component eigenvectors of the covariance matrix of the
zero-mean-adjusted samples of spectrum are also used to extract the sound signature as in [20].
Some vehicle acoustic signatures have a pattern of relation between the harmonics amplitude.
Harmonics are the peaks of the spectral domain. The relation between the amplitude and the
phase of these peaks is used to form the feature vector. Harmonic Line Association (HLA) feature
vector is used in [30], where the magnitude of the second through 12th harmonic frequency
components are considered as the feature vector to be used for vehicle classification. Different
algorithms are used to estimate the fundamental frequency. In [36], two sets of features are
extracted from the vehicle sound. The first one is based on the harmonic vector. The second one
is a key frequency feature vector. In [37], the number of harmonics is modeled as a function of

the vehicle type. Looking for stable features other than the harmonics relation, Ref. [50] models
the vehicle acoustic signature by a coupled harmonic signal. Cepstral coefficients (CC) are the
coefficients of the inverse Fourier Transform of the log of the magnitude of the spectrum [23].
Mel-frequency cepstral coefficients (MFCC) is used in [18],[40],[43] as a feature extractor, where
the feature vector is made up of few of the lowest cepstrum coefficients. Mel-frequency cepstrum
(MFC) is a representation of the short-term power spectrum of a sound, where the log power
spectrum on a nonlinear mel scale of frequency is transformed based on a linear cosine
transform.
Two types of spectral features are explored in [21]: Non-parametric FFT-based PSD estimates,
and Parametric PSD estimates using autoregressive (AR) modeling of time series. In AR model
the value of any variable at any time is modeled as a function of a finite number of the past values
of that variable. The number of the involved past values is called the model order. AR of the first
order is a Markov chain, where the current value depends only on the previous value. A randam
variable X can be modeled at time t using AR of order P as follows:
tktk
p
k
t xx ωθ +−∑1=
=
(1)
where kθ denotes the corresponding autoregressive coefficients. tω is a white gaussian noise
with zero mean. If kθ is varying with time then the AR process is called Time Varying
Autoregressive (TVAR). TVAR is used to model the acoustic signal in [26, 51]. A filter bank is
used based on the biology based hearing model (BHM) as a feature extraction system in [27].
Feature Extraction Example using the Spectrum Distribution:
The goal is to develop a scheme for extracting a low dimension feature vector, which is able to
produce good classification results. The first feature extraction technique of acoustic signals in
this paper is based on the low frequency band of the overall spectrum distribution. The low
frequency band is utilized, because most of the vehicle's sounds come from the rotating parts,
which rotate and reciprocate in a low frequency, mainly less than 600 Hz as it is clear in Fig. 6.
Sounds of moving ground vehicles are recorded at the nodes at a rate of 4960 Hz as in Fig.5.
After the positive detection decision, a signal of event is preprocessed as the following:
Figure 5: Time domain for three different sounds for two different vehicles v1 and v1.

Figure 6: Frequency distribution for three different sounds for two different vehicles v1 and v1. Fs
is sampling frequency =4960. Fw is FFT window size=512.
DC bias should be removed by subtracting the mean from the time series samples.
)(*
1
)(=)(
1=
nx
N
nxnx i
N
n
ii ∑−
(2)
.
Feature vector will be the median of the magnitude of the STFT of a signal of event. It will be
computed as the following: the magnitude of the spectrum is computed by FFT for a hamming
window of size 512, without overlapping.
))((=)( nxFFTWX ii (3)
After this, the spectrum magnitude is normalized for every frame
)(
)(
=)(
1=
WX
WX
WX
i
K
W
i
i
∑
(4)
where K is the window size. The median of all frames is considered as the extracted feature
vector.
))((=)( WXmedianWX iif
(5)
The mean of all frames could also be considered as the extracted feature vector.
)(
1
=)(
1=
WX
Z
WX i
Z
i
if ∑
(6)
where kNz /= . The first 64 points of the median of the spectrum magnitude contain up to 620
Hz. This gives a 64 dimensional vector that characterizes each vehicle sound. We compared
feature extraction using the mean and the median. The median gives better results, specially for
noisy environments. Fig.7 displays the acoustic spectral distribution of vehicle 1 and vehicle 2.
For the unknown utterance, the same steps are done, except one frame of FFT is considered as
the feature to be classified to reduce the computational cost, because this FFT computation is
performed online. This can be extended to have multiple frames, but this will increase the
computational cost.

Figure 7: Acoustic spectra distribution of vehicle 1 and vehicle 2. To the left is vehicle 1
2.3 Time Frequency Domain
Short Time Fourier Transform (STFT) is used in [38] to transform the overlapped acoustic
Hamming windowed frames to a feature vector. Ref. [52] proposes a probabilistic classifier that is
trained on the principal components subspace of the short-time Fourier transform of the acoustic
signature. Wavelet transforms provide multi-resolution time-frequency analysis [53]. Wavelet
transforms (WT) is the the projection of a signal onto the wavelet. Wavelet is a series of functions
)(, tbaψ
derived from a base function )(tψ by translation and dilation.
)(
||
1
=)(,
a
pt
a
tba
−
ψψ
(7)
where a is called scale parameter, b is called translation or shift parameter, and )(tψ is called
wavelet base function. Wavelet Transform is called CWT when values of a and b are continuous,
and it is called DWT when they are discrete [54]. Discrete Wavelet Transform (DWT)
approximation coefficients y are calculated by passing the time series samples x through a low
pass filter with impulse response g.
)(g)(x=)(g*)(x=)(y
=
knknnn
k
−∑
∞
−∞ .
The signal is also decomposed simultaneously using a high-pass filter h. The outputs from the
high-pass filter are the detail coefficients. The two filters are related to each other. DWT is exactly
the same as the Wavelet Packet Transform (WPT) except that in DWT the next level is the result
of one step of wavelet transform of the approximation branch and not the detail one. Wavelet
packet transform can be viewed as a tree structure. The root of the tree is the time series of the
vehicle sound. The next level is the result of one step of wavelet transform. Subsequent levels in
the tree are obtained by applying the wavelet transform to the low and high pass filter results of
the previous step's wavelet transform. The Branches of the tree are the blocks of coefficients.
Each block represents a band of frequency. Feature extraction of acoustic signals is based on the
energy distribution of the block coefficients of wavelet packet transform. A wavelet-based
acoustic signal analysis of military vehicles is presented in [13, 55]. Discrete Wavelet Transform

(DWT) is used in [13] and [14] to extract features using statistical parameters and energy content
of the wavelet coefficients. Wavelet Packet Transform (WPT) has a higher frequency resolution
than the DWT [56]. WPT is also used to extract vehicle acoustic signatures by obtaining the
distribution of the energies among blocks of wavelet packet coefficients like in [24],[39]. Ref. [53]
has a proof that wavelet analysis methods is suitable for feature extraction of acoustic signals
.
Feature Extraction Using Wavelet Packet Transform:
After the positive detection decision, a one second time series is preprocessed as the following:
• The wavelet packet transform is applied for this signal then the energy of each block
coefficients of the (L) level is calculated.
• This approach provides a vector of length = original time series length
L
/2 . Which is
considered the feature vector.
Figure 8: Wavelet block energy distribution for vehicle one in first row for three different
sounds and for vehicle 2 in the second row.
Fig. 8 displays the blocks energy distribution for vehicle 1 and vehicle 2. In this paper we used
classification rate as the metric for the evaluation of the feature extraction performance. But this
metric depends on the classifier itself. Thus, we compare the classification rate for two classifiers
as shown in Fig. 12.
2.4 Feature Extraction Performance Using Separability Measures
Separability measures provide a measure of class discriminability based on feature space
partitioning. A good feature vector extractor provides close feature vectors for the same class,
and far feature vectors for distinct classes. The goal is to have a feature extraction method that
has high distance between distinct classes and low distance within each class. The metric, in this
paper, is the separability ratio (sr), which is the ratio between the intraclass distance and the
average interclass distance [57].
l
g
D
D
sr =
(8)
2
1
1=1=
]))([(= T
iikiik
in
ki
i
C
i
g mVmV
n
P
D −−∑∑
(9)

gD
represents the average of the variances of distance within all classes. ikV
is the normalized
feature vector. C is the number of classes. iP
is the probability of class i. in
number of vectors in
class i. im is the mean vector for class i.
2
1
1=
]))([(= T
iii
C
i
l mmmmPD −−∑
(10)
lD
represents the average of the distances between all classes. m is the mean for all classes.
i
ik
in
k
i
n
V
m
∑1=
=
(11)
i
ik
in
k
C
i
n
V
m
∑∑ 1=1=
=
(12)
The smaller the ratio is the better the separability. This means that the best feature extraction
scheme is the one that decreases gD
and increases lD .
3. CLASSIFICATION TECHNEQUES
Classifiers provide the functions or the rules that divide the feature space into regions, where
each region corresponds to a certain class. Having a number of N -dimensional feature vectors
for each class, a function can be deduced to partition the N feature space to number of regions,
where each region represents a class. This process is called supervised learning or classification.
Classifiers can be categorized to parametric or non-parametric. Some researches combine
multiple different classifiers, called compound classifiers.
3.1 Parametric Classifiers
Parametric classifiers are the classifiers that can be represented in closed-form. for instance,
assuming that the distribution of a certain class as a parametric form such as Gaussian. Some
classifiers are based on discrimination function with a certain parametric form such as support
vector machine. Below are the most parametric classifiers that have been used in vehicle
classification based on acoustic signature.
3.1.1 Bayesian Classifier
Bayesian classifier is a probabilistic classifier based on using Bayes' theorem. Maximum
likelihood (ML) is used to estimate the Bayesian classifier parameters. Maximum A posteriori
Probability (MAP) can also be considered as a generalization of ML. Each class is assumed to be
independent instances of parametric distributed random process. A naive Bayes classifier is a
variation of the Bayesian classifier with assumption of an independent feature model. Bayesian
classifier is used in many research papers with assumption that each class is a normal distributed
random process [35],[41],[58],[59] where features vector S of each class C is assumed to be
independent instances of normally distributed random process.
),()|(p iii NS Σ≈ µθ (13)

i=1,2 ....C. iµ and iΣ are the mean and covariance matrix respectively. iθ is the parameter set
of
th
i distribution, iθ = ii Σ,{µ }.
)]()(
2
1
[
||)(2
1
=)|(p 1
1/2/2
µµ
π
θ −Σ−−
Σ
−
SSexpS H
di
(14)
To represent each training set of each class as a distribution with Σˆ and µˆ parameters the
likelihood of θˆ
)|(=)(
1=
θθ i
n
k
i slnPl ∑
(15)
should be maximized by equating ilθ∆ =0, then ML estimations of µ and Σ are
k
n
k
s
n
∑1=
1
=ˆµ
(16)
H
kk
n
k
ss
n
)ˆ)(ˆ(
1
=ˆ
1=
µµ −−Σ ∑
(17)
for minimum error classification the a posteriori probability should be maximized
)(p)|(p
)(p)|(p
=)|(p
1=
jj
C
j
ii
i
S
S
S
θθ
θθ
θ
∑
(18)
)(xhi denote the logarithmic version of
)|(p Siθ
GlnSlnxh iii −+ )(p)|(p=)( θθ (19)
where G is constant can be ignored in the optimization
DSSxh i
H
i −−Σ−− −
)()(
2
1
=)( 1
µµ
(20)
where D is constant then any vehicle feature vector x is classified to class i according to the
discriminant functions
)(xhi if
)(>)( xhxh ji
for all
ij ≠ . Linear discriminant Analysis (LDA)
assumes that the class covariances iΣ
are identical. LDA is used as a linear classifier in [24]
ML is the optimum classifier but it needs large number of training set. Training ML classifier
with small number of training set will not give an invertible covariance matrix. This makes it hard
to compute the discriminant functions.

3.1.2 Support Vector Machine (SVM)
SVM is widely used as a learning algorithm for classifications and regressions. SVM classify data
ix by class label
1}1,{ −+∈iy given a set of examples
},{ ii yx by finding a hyperplane bwx+
,
n
Rx∈ which separate the data point ix of each class . as in Fig. 9.
Figure 9: An Example of Two Classes Problem. Squares and Circles Represent Class 1 and
Class 2 Respectively.
)(=)( bwxsignxg + (21)
where w is the weight vector, b is the bias. SVM choose the hyperplane that maximize the
distance between the hyperplane and the closest points in each feature space region which are
called support vectors. So the unique optimal hyperplane is the plane that maximize this distance
PPwbwxi /|| + (22)
This is equivalent to the following optimization problem
1)(..,
2
2
, ≥+ bxwyts
w
min i
T
ibw
PP
(23)
For the cases that nonlinear separable, a kernel function maps the input vectors to a higher
dimension space in which a linear hyperplane can be used to separate inputs. So the
classification decision function becomes:
)),(( 0
bppKysign iii
SVsi
+∑∈
α
(24)
where SVs are the support vectors.
0
iα and b are a lagrangian expression parameters.
pK( ,
)ip is the kernel function. It is required to represent data as a vector of a real number to
use SVM to classify moving ground vehicles. Performance of SVM classifier for vehicle
acoustic signature classification for both feature extraction methods is also evaluated. It is found
that SVM is a good classifier for stochastic signal in WSN [12], [42], [60]–[63].

3.1.3 Gaussian Mixture Model (GMM)
Due to the constraints in WSN resources, parametric models such as Gaussian mixture model is
preferred to non-parametric models [43]. Modeling of acoustic signal in WSN using a parametric
model, like GMM requires little resources, and has a good pattern matching performance [43].
GMM is a statistical method that is used for classification and clustering. GMM is a linear
combination of M-Gaussian pdfs. Let x be a N-dimensional feature vector, then the distribution of
x is as follows:
);(=)(
1=
ii
m
i
m xxf θφα∑
(25)
where
1=1= i
m
i
α∑ ,
mii 1,...,:0 ∈≥α
iα is the mixing weight,
);( ix θφ is the Gaussian mixture component. Component i has N-
variate Gaussian density function with weight iα , mean vector iµ , and covariance matrix iΣ .
Expectation maximization (EM) is one of the common algorithm that is used to obtain the GMM
parameters
),,(= iiii ΣµΦ α from the training set. The GMM generated from the training set will
be used in vehicle classification as in Fig. 10.
Figure 10: Main block diagram of pattern recognition
Any vehicle feature vector x is classified to class iC if it maximizes
)().|(=)|( iii CpCxpxCp . If all the classes are assumed to occur with the same probability,
then the concern is to maximize
)|( iCxp for every possible class. GMM is used as a classifier
in WSN based on the features that are extracted from the vehicle sounds in [43]. Ref. [43]
concludes that the GMM, as a parametric classifier, outperforms the KNN and SVM classifiers,
and it also concludes that GMM needs relatively less resources.
3.1.4 Hidden Markov Model (HMM)
Acoustic signals could be modeled as HMM. HMM has a specific discrete number of unobserved
states; each state has a transition probability to any other state and an initial probability. Each
state may be considered as representing a certain sound of the vehicle [23]. Ref. [23] models the
cepstral coefficients that are obtained from the time domain signal as HMM, where the pdfs of the
states are assumed to be Gaussian with non-zero means and with a diagonal covariance matrix.
Modeling the vehicle sounds as HMM is based on the assumption that the acoustic signal of the
vehicle is consisting of a sequence of a discrete number of sounds, where the statics of each
sound of these sounds is described by a separate state. The parameters of the HMM are: the
state transition probability to any other state, the initial probability for each state, and the

observation pdf parameters for each state. Estimation of the maximum likelihood parameters of
the HMM given a data set of the vehicle sounds can be done by a special case of the
Expectation-maximization algorithm called the Baum-Welch algorithm; it is also known as the
forward-backward algorithm. HMM implementation for vehicle classification is based on the
estimation of the sequence of states, given a sequence of observations. Some known algorithms
are used for that such as the Viterbi algorithm. GMM is static pattern model, while HMM is a
sequential pattern model.
3.1.5 Minimum Distance Approach
Minimum Distance Approach (MPP) is a simple parametric classifier that is based on the
minimum distance between the feature vector and the average of the class distribution. MPP
assume the distribution of the training set of each class as Gaussian distribution. MPP measure
the distance between the average of each class distribution and the feature vector of the test
data, then it corresponds the test data to the class that has the minimum distance. MPP is used
as a classifier in Refs [13],[19],[44] .
3.2 Non-parametric Classifiers
In this kind of classifiers no assumptions are made about the probability density function for each
class.
3.2.1 KNN classifier
KNN is a simple and accurate method for classifying objects based on the majority of the closest
training examples in the feature space. KNN is rarely used in wireless sensor networks because it
needs large memory and high computation. In our experiments we set K to be three. So the KNN
classifier finds the closest three neighbors out of all the training set. After that, the KNN classifier
counts how many of these three is in class one and how many is in class two then based on the
majority classify the tested one. KNN is implemented in many literatures as a benchmark to
evaluate other classifiers [21],[31],[34],[35],[44]
3.2.2 Artificial Neural Network (ANN)
Artificial neural networks are a learning intelligent tools used to solve problem that hard to be
modeled analytically. A key feature of neural networks is the ability to learn from examples of
input/ output pairs in a supervised fashion. In [40], rough neural network is used to recognize type
of vehicles in WSN. The network has 25 input neurons, corresponding to the 25-dimisional
feature vector, 25 hidden layer neurons, and 4 output neurons. Neural network classifier and the
maximum likelihood classifier are compared concerning their advantages and disadvantages in
[45]. Artificial neural networks were used as a technique to classify and identify vehicles based on
acoustic signals in [46], where a two-layer back propagation neural network is trained by the
output of a self organized maps neural network.
3.2.3 Fuzzy Logic Rule-Based classifier
Fuzzy Logic Rule-Based classifier (FLRBC) maps the input to the output based on some rules
[64]. Fuzzy logic inference is a simple approach to solving a problem rather than attempting to
model it mathematically. Empirically, the fuzzy logic inference depends on human's experience
more than the technical understanding of the problem. As in Fig. 11, fuzzy logic inference
consists of three stages:
1. Fuzzification: map any input to a degree of membership in one or more membership
functions, the input variable is evaluated in term of the linguistic condition.
2. Fuzzy inference: fuzzy inference is the calculation of the fuzzy output.
3. Defuzzification: defuzzification is to convert the fuzzy output to a crisp output.

Figure 11: The stages of fuzzy logic inference design.
Cascaded fuzzy classifier is proposed in [47] to classify ground moving vehicles locally at sensor
nodes in WSN.
3.2.4 Decision Tree
Decision tree is a nonlinear classifier that depend on a multistage decision system, where the
classes are sequentially rejected until reach the accepted class. This kind of classifier split the
feature space into unique regions, where each region represents a class [65]. In Refs.
[38],[43],[66] decision tree is utilized as a classifier. C4.5 algorithm is used to generate a decision
tree in [38]. Decision tree is sometimes describes as classification tree or regression tree.
Classification And Regression Tree (CART) analysis is used to refer to both of classification and
regression.
3.3 Classifiers Evaluation and Comparison
Classification evaluation is to measure how accurate the classifier is. There are three main
classification metrics that are used to evaluate the performance of the classifier namely: Correct
classification rate, detection probability, and false alarm probability. Correct classification rate is
the ratio of the correctly classified events from all the samples for all classes. Detection
probability for a certain class is the ratio of the correctly classified events from the samples of that
class. False alarm probability for a certain class is the ratio of the wrongly classified events for
that class from all samples of other classes. Classifiers might have a high accuracy if tested with
the same training data. Thus, a classification validation is needed. There are two common
methods that are used for classification validation. The first is called hold-out or split-sample
method, where a single part of the data is set a side for testing. The second method is the cross
validation method. k-fold cross-validation divides the data into k subsets. The classifier is trained
k times, each time leaving out one of the subsets from training to be used for testing the classifier.
If k equals the sample size, the method is called leave-one-out. Ref [45] compared the
recognition rate and the robustness of two classifiers, neural network classifier and maximum
likelihood classifier. Neural fuzzy techniques for classification of vehicle acoustic signal is used in
Ref. [29]. In [38], Four different classifiers decision tree (C4.5), KNN, probabilistic neural network
(PNN) and SVM are cpmpared. The classification results indicate the performance of SVM
outperforms C4.5, KNN, and PNN. SVM, ML, and KNN are used in [35] to evaluate their data set.
In this paper, five different classifiers are compared as in Fig. 12.

Figure 12: A comparison of the correct classification rates for different classifiers for two different
feature extraction methods. The first one is base on the spectrum analysis and the other one is
based on wavelet packet transform.
3.4 Single Target Classification
Single vehicle classification is to label or classify a vehicle to one of the predefined classes.
Classification is based on the feature vector, which is generated from the observed continuous
stochastic signal. This stochastic signal could be acoustic, seismic, electromagnetic, or any other
kind of signals. The feature vector will be the input of the classifier. The classifier could be any
kind of the classifiers that are mentioned in the classifiers section. Classifiers could be parametric
or non parametric. The parameters of the parametric one are estimated from a training set. Non
parametric classifiers are also trained by a training set. Single vehicle classification techniques
can be used for multiple vehicle classification after sources separation or with the assumption that
that the sensor will not observe two or more vehicles at the same time. This assumption is not
realistic, especially in battlefield scenarios.
3.5 Multiple Targets Classification
Many researchers assume that multiple targets could be separated in time and space. However,
in many of the surveillance applications this assumption is not realistic, where multiple targets can
exist in the sensing range of one sensor at the same time as in Fig. 13. This makes the sensor
observes a combination of signals. The combination of multiple signals can be modeled as linear,
nonlinear, or convoluting combination of the single target signals. Ref. [67] exploits the classifier
that is trained in single target classification to classify a convoy of vehicles. Most of the literature
models the multiple targets classification problem as a Blind Source Separation (BSS) problem.
BSS problem has been tackled in the literature in many ways, such as neural network [68]-[70]
and Independent Component Analysis (ICA). ICA is frequently used to separate or extract the
source signals [71]-[75]. In [76] the source extraction problem in wireless sensor network is
studied in two different sensor network models. Fast fixed-point ICA algorithm is used for source
separation [51] presents a statistical method based on particle filtering for the multiple vehicle
acoustic signals separation problem in wireless sensor networks. In [77], a recognition system
links BSS algorithm with an acoustic signature recognizer based on missing feature theory (MFT).
The result of the comparison between FasICA, Robust SOBI, and their work shows that both of
former algorithms are better for mixtures of two signals and more. Refs. [78],[79] discuss problem
of source estimation in sensor network for multiple targets detection, then a distributed source
estimation algorithm is developed. These solutions have some drawbacks that make it hard to be
implemented in WSNs. It is evident that the manager nodes need to perform source separation
and source number estimation. These tasks are computationally intensive when executed on the
individual sensor nodes. The manager node does the following: estimation of the number of
sources, separation or extraction of sources, classification of sources. [80] presents a system that
is able to recover speech signal in the presence of additive non-stationary noise. This done

through a combination of the classification and mask estimation. Ref. [37] uses a multi-variate
Gaussian classifier for classifying individual acoustic targets after beamforming the received
signal in the direction of the targets. We direct the reader for more information in beamforming to
[81]. Classification of multiple targets without the separation of the sources based on multiple
hypothesis testing is an efficient way of classification [82]. A distributed classifiers based on
modeling each target as a zero mean stationary Gaussian random process and the same for the
mixed signals is proposed in Ref. [83].
Figure 13: A summary diagram of the techniques that are used in multiple vehicle classification
using WSNs.
4. COLLABORATIVE CLAASSIFICATION
Data fusion, information fusion, data aggregation, multisensor integration and sensor fusion are
terms that have been used interchangeably in the literature to refer to the idea and theory of
fusion. In WSNs data fusion is needed for the following reasons: sensor failure, sensor and
technology imprecision, limitations in spatial and temporal coverage [84]. Information fusion can
be classified according to the relationships between the sources as: redundant, cooperative , or
complementary. It also can be classified according to the abstraction level of the processed data
as: low-level (raw data), medium-level, high-level (decision fusion), or multi-level fusion.
The main objective of collaboration classification is to extract the most beneficial information from
the collected data. Because every target has its own signature according to the type of generated
signal. Deployment of different kinds of sensors, in the same sensor node or in different sensor
nodes, increases the performance of collaborative signal processing. This stems from the fact
that every sensor type has a different interference and a measurements noise. Efficient and
reliable decision making needs data fusion and collaborative signal processing. Distributed
classification algorithms fuse signal or decisions from multiple sensor nodes, then classify the
targets based on a priori statistical information [85],[86]. Collaboration could be across the sensor
nodes, or within a sensor node only when it includes multiple modalities of data. Ref. [67] shows
the improvement in classification error because of the collaboration. WSNs have two kinds of
collaboration signal processing models: data fusion and decision fusion. For more information in
data and decision fusion see [87],[88]. Ref. [89] analyzes both models. Exploiting the
collaboration among the sensor nodes enhances even the sensing coverage for the network [90].
Decision fusion has less accuracy than data fusion. However, data fusion has more computation
and communication overhead than the data fusion. In decision Fusion, Sensor nodes do some
processing then send the decision to the manager node as in Fig. 15, where these decisions
could be hard or soft decisions [91]-[93]. Manager node fuses all the decisions and come up with
the best decision. Rules of decision fusion could be based on: voting, averaging, Bayesian,
Dempster-Shafer (DS) sum or other rules as in [94]. An example of decision fusion is the tracking
system that is proposed in [95], where detection and classification are performed in the sensor

node while tracking is performed in the sink node. Data and decision fusion are increasingly
implemented in sensor network because of hardware improvement, advances in processing
methods, and advances in sensor technology [96]. Fig 14 shows some of of the data fusion
applications. Data and decision fusions techniques answer the question of how to combine the
data and decisions from the sensor nodes in a network to obtain a final decision with optimal
resource consumption. Sensor nodes make the measurements, then send row measurements,
processed measurements, or local posterior to the fusion center. Fusion architecture can be
hierarchical or distributed. In hierarchical fusion, the data or decision is fused from the sensor
node to the higher level fusion center. While in distributed fusion architecture, the data or decision
is broadcasted to all other fusion centers.
There are various scenarios of data and decision fusion for single and multiple sensors within the
sensor node or cross over the network. Ref. [97] has a survey that focuses on the decision fusion.
Ref. [98] studies a distributed decision fusion, where the local decision makers send the ranking
of the hypotheses, rather than just the most likely hypothesis. A consensus algorithm which
weighs the measurements of the neighboring nodes is presented in [99].
Data fusion from seismic and acoustic signal improves the classification accuracy [67]. In Ref.
[43], a decision tree generated by the classification and regression tree algorithm is used to fuse
the information from heterogeneous sensors. Multi-resolution integration (MRI) algorithm is used
in [19] as a data fusion algorithm to handle the uncertainty of sensor output. MRI algorithm is
based on building an overlap function from the outputs of the sensors, where this function is
resolved at finer scales.
5. OPEN REASEARCH
For single node processing, researcher use mathematical or statistical tools to extract the
features that can represent a unique signature for each vehicle or class. Then another tools are
used to classify any new utterance to one of the predefined classes. The main goals of this kind
of research is either to increase the performance of the classification or decrease the complexity.
The relation between performance and complexity is a trade-off relation. Thus the research is
open in both areas. Optimal classifier is the classifier that increase the performance and decrease
the complexity. Thus, it will be so beneficial to have a standard metrics that have both measures
and have a public data base where researcher can evaluate their algorithms based on these
metrics using a public data base. Single node processing versus collaborative processing is a hot
area of research. Collaborative processing answers the question of how to combine the data and
decisions from the sensor nodes in a network to obtain a final decision with optimal resource
consumption. Collaborative processing is an open area for research. Fusion modeling, where the
signal to noise ratio for both acoustic and communication channel are considered, is critical to
find the optimal number of decision makers. Utilization of time and spatial correlation across
different nodes is crucial in collaborative processing. All of the above is related to the signal
processing techniques that are used for feature extraction, classification, and data and decision
fusion. However this is related to the communication protocols that are used in WSN.

Figure 14: Data fusion applications.
6. CONCLUSIONS
The recent research related to classification process of ground vehicles in wireless sensor
network, based on acoustic signals, is reviewed in this paper. Classification process involves
three main components: feature extraction, classifier design and selection, and information
fusion. Feature extraction methods are classified based on the extraction domain to time,
frequency, and time/frequency. Classifiers are also categorized into parametric and non
parametric classifiers. In wireless sensor networks decision fusion is preferred rather than data
fusion because of the power constraints. This paper proposed two feature extraction methods.
One is based on wavelet packet transform and the other is based on spectrum distribution. both
methods give almost the same separability and classification rate.
Figure 15: One cluster of a wireless sensor network.

3. REFERENCES
[1] K. Ro¨ mer, F. Mattern. “The design space of wireless sensor networks”. IEEE Wireless
Communications, 11(6)54–61, 2004
[2] T. Arampatzis, J. Lygeros, S. Manesis. “A survey of applications of wireless sensors and
wireless sensor networks”. In Intelligent Control, Proceedings of the 2005 IEEE International
Symposium on, Mediterrean Conference on Control and Automation, 2005
[3] R. Tan, G. Xing, J. Wang, and H. C. So. “Exploiting reactive collaborative target
detection in wireless sensor networks”. IEEE Transactions on Mobile Computing, 99(1):
5555.
[4] L. B. Saad and B. Tourancheau. “Multiple mobile sinks positioning in wireless sensor
networks for buildings”. In Sensor Technologies and Applications, 2009. SENSORCOMM
’09. Third International Conference, 2009
[5] R. Zurawski. “Keynote: Wireless sensor network in industrial automation”. In Embedded
Software and Systems, ICESS ’09. International Conference, 2009
[6] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci. “Wireless sensor
networks: a survey”. Computer Networks, 38(4):393–422, 2002. [Online]. Available at:
http://www.sciencedirect.com/science/article/B6VRG-44W46D4-
1/2/f18cba34a1b0407e24e97fa7918cdfdc
[7] P. Gajbhiye and A. Mahajan. “A survey of architecture and node deployment in wireless
sensor network”. In Applications of Digital Information and Web Technologies, ICADIW,
First International Conference, 2008
[8] T. He, S. Krishnamurthy, J. A. Stankovic, T. Abdelzaher, L. Luo, R. Stoleru, T. Yan, L.
Gu, J. Hui, and B. Krogh. “Energy-efficient surveillance system using wireless sensor
networks”. In MobiSys ’04: Proceedings of the 2nd international conference on Mobile
systems, applications, and services. New York, NY, USA: ACM, 2004
[9] T. Canli, M. Terwilliger, A. Gupta, and A. Khokhar. “Power - time optimal algorithm for
computing fft over sensor networks”. In SenSys ’04: Proceedings of the 2nd international
conference on Embedded networked sensor systems. New York, NY, USA: ACM, 2004
[10] C. F. Chiasserini and R. R. Rao. “On the concept of distributed digital signal processing
in wireless sensor networks”. In Proceedings of MILCOM, 2002
[11] D. Estrin, L. Girod, G. Pottie, M. Srivastava “Instrumenting the world with wireless
sensor networks”. In 2001 IEEE International Conference on Acoustics, Speech, and Signal
Processing. Proceedings (Cat. No.01CH37221), 4: 2033–6, 2001
[12] G. P. Mazarakis and J. N. Avaritsiotis. “Vehicle classification in sensor networks using
time-domain signal processing and neural networks”. Microprocess. Microsyst., 31(6) 381–
392, 2007
[13] H. Choe, R. Karlsen, G. Gerhart, and T. Meitzler. “Wavelet-based ground vehicle
recognition using acoustic signals”. In Proceedings of the SPIE - The International Society
for Optical Engineering, Conference Paper, Wavelet Applications III, Orlando, FL, USA,
2762: 434–45, SPIE, 1996

[14] A. Khandoker, D. Lai, R. Begg, M. Palaniswami. “Wavelet-based feature extraction for
support vector machines for screening balance impairments in the elderly”. Neural
Systems and Rehabilitation Engineering, IEEE Transactions on, 15(4)587–597, 2007
[15] M. C. Wellman, N. Srour and D. B. Hillis. “Feature extraction and fusion of acoustic and
seismic sensors for target identification”. In Society of Photo-Optical Instrumentation
Engineers (SPIE) Conference Series, ser. Society of Photo-Optical Instrumentation
Engineers (SPIE) Conference Series, G. Yonas, Ed., 3081:139–145, 1997
[16] G. Succi, T. Pedersen, R. Gampert and G. Prado. “Acoustic target tracking and target
identification-recent results”. In Proceedings of the SPIE - The International Society for
Optical Engineering, 3713:10–21, 1999
[17] M. E. Hohil, J. R. Heberley, J. Chang, A. Rotolo. “Vehicle counting and classification
algorithms for unattended ground sensors”. E. M. Carapezza, Ed., SPIE, 5090(1):99–110,
2003. Available: http://link.aip.org/link/?PSI/5090/99/1.
[18] M. E. Munich. “Bayesian subspace methods for acoustic signature recognition of
vehicles”. In Proc.EUSIPCO, 2004
[19] X. Wang H. Qi. “Acoustic target classification using distributed sensor arrays”. In Proc.
IEEE ICASSP, 4:4186–4189, 2002
[20] H. Wu, M. Siegel, P. Khosla. “Vehicle sound signature recognition by frequency
vector principal component analysis”. Instrumentation and Measurement, IEEE
Transactions on, 48(5):1005–1009, 1999
[21] D. Li, K. D. Wong, Y. H. Hu, and A. M. Sayeed. “Detection, classification and tracking
of targets in distributed sensor networks”. In IEEE Signal Processing Magazine, 2002
[22] H. Maciejewski, J. Mazurkiewicz, K. Skowron, and T. Walkowiak. “Neural networks for
vehicle recognition”. In Proc. 6th Conference on Microelectroncs for Neural Networks,
Evolutionary and Fuzzy System, 1997
[23] W. J. Roberts, H. W. Sabrin, and Y. Ephraim. “Ground vehicle classification using
hidden markov models”. In Atlantic coast technologies Inc., Silver Spring MD, 2001
[24] A. Averbuch, E. Hulata, V. Zheludev, and I. Kozlov. “A wavelet packet algorithm for
classification and detection of moving vehicles”. Multidimensional Syst. Signal Process,
12(1):9–31, 2001
[25] J. E. Lopez, H. H. Chen, and J. Saulnier. “Target identification using wavelet-based
feature extraction and neural network classifiers”. in CYTEL SYSTEMS INC HUDSON MA,
1999
[26] K. B. Eom. “Analysis of acoustic signatures from moving vehicles using time-varying
autoregressive models”. Multidimensional Syst. Signal Process, 10(4):357–378, 1999
[27] L. Liu. “Ground vehicle acoustic signal processing based on biological hearing models”.
Master’s thesis, University of Maryland, 1999
[28] N. B. Thammakhoune and S. W. Lang. “Long range acoustic classification”. Sanders a
Lockheed Martin Company, Tech. Rep., 1999
[29] S. Somkiat “Neural fuzzy techniques in vehicle acoustic signal classification”. Ph.D.

dissertation, chair-Vanlandingham, Hugh F. 1997
[30] A. Y. Nooralahiyan, H. R. Kirby, D. McKeown. “Vehicle classification by acoustic
signature”. Mathematical and computer modeling, 27:9 –11, 1998
[31] M. Baljeet, N. Ioanis, H. Janelle. “Distributed classification of acoustic targets in
wireless audio-sensor networks”. Comput. Netw., 52(13):2582–2593, 2008
[32] L. Chun-Ting, H. Hong, F. Tao, L. De-Ren, S. Xiao. “Classification fusion in wireless
sensor networks”. ACTA AUTOMATICA SINICA, 32(6):948–955, 2006
[33] S. S. Yang, Y. G. Kim1, H. Choi. “Vehicle identification using discrete spectrums in
wireless sensor networks”. Journal Of Networks, 3(4):51–63, 2008
[34] B. Malhotra, I. Nikolaidis, and M. Nascimento. “Distributed and efficient classifiers for
wireless audio-sensor networks”. In Networked Sensing Systems, 2008. INSS 2008. 5th
International Conference on, 2008
[35] M. F. Duarte, Y. H. Hu. “Vehicle classification in distributed sensor networks”. Journal
of Parallel and Distributed Computing, 64(7):826–838, 2004, computing and
Communication in Distributed Sensor Networks. Available at:
http://www.sciencedirect.com/science/article/B6WKJ-4CXD0JJ-
1/2/64f671263463155e2afd7bf778c3a7dd
[36] B. Guo, M. Nixon, and T. Damarla. “Acoustic information fusion for ground vehicle
classification”. In Information Fusion, 2008 11th
International Conference, 2008
[37] R. D. T., P. Tien, and L. Douglas. “An algorithm for classifying multiple targets
using acoustic signatures”, 2004
[38] H. Xiao1, Q. Yuan1, X. Liu1, and Y. Wen. “Advanced Intelligent Computing Theories
and Application, with Aspects of Theoretical and Methodological Issue.” Springer Berlin /
Heidelberg, 2007
[39] A. Amir, Z. V. A., R. Neta, S. Alon “Wavelet-based acoustic detection of moving
vehicles”. Multidimensional Syst. Signal Process., 20(1):55–80, 2009
[40] H. Qi, X. Tao, and L. H. Tao. “Vehicle classification in wireless sensor networks based
on rough neural network”. In ACST’06: Proceedings of the 2nd IASTED international
conference on Advances in computer science and technology. Anaheim, CA, USA: ACTA
Press, 2006
[41] D. Li, K. Wong, Y. H. Hu, and A. Sayeed. “Detection, classification, and tracking of
targets”. Signal Processing Magazine, IEEE, 19(2):17–29, 2002
[42] Q. Xiao-xuan, J. Jian-wei, H. Xiao-wei and Y. Zhong-hu. “An approach of passive
vehicle type recognition by acoustic signal based on svm”. 2009
[43] Y. Kim, S. Jeong, D. Kim, T. S. Lo´ pez. “An efficient scheme of target classification
and information fusion in wireless sensor networks”. Personal Ubiquitous Comput., 13(7):
499–508, 2009
[44] Y. Sun and H. Qi. “Dynamic target classification in wireless sensor networks”. In
Pattern Recognition, ICPR ,19
th

[45] W. Duan, M. He, Y. Chang and Yan Feng. “Acoustic objective recognition in wireless
sensor networks”. In 2009 4th IEEE Conference on Industrial Electronics and Applications,
2009
[46] R. Mgaya, S. Zein-Sabatto, A. Shirkhodaie, and W. Chen. “Vehicle identifications using
acoustic sensing”. In SoutheastCon, 2007 Proceedings. IEEE, 2007
[47] P. Qiang, W. Jianming, C. Hongbing, L. Na, and L. Haitao. “Improved ds acoustic-
seismic modality fusion for ground-moving target classification in wireless sensor
networks”. Pattern Recogn. Lett., 28(16):2419–2426, 2007
[48] S. Erb. “Classification of vehicles based on acoustic features”. Master’s thesis,
Begutachter: Univ.-Prof. Dipl.-Ing. Dr. Bernhard Rinner, 2007
[49] A. Aljaafreh and L. Dong. “An evaluation of feature extraction methods for vehicle
classification based on acoustic signals”. In Networking, Sensing and Control (ICNSC),
2010 International Conference, 2010
[50] D. Lake. “Harmonic phase coupling for battlefield acoustic target identification”. In
Acoustics, Speech and Signal Processing, 1998. Proceedings of the 1998 IEEE
International Conference on, 4:2049–2052, 1998
[51] Y. Kaia, H. Qia, W. Jianminga, L. Haitao. “Multiple vehicle signals separation based on
particle filtering in wireless sensor network”. Journal of Systems Engineering and
Electronics, 19(3):440–446, 2008
[52] M. E. Munich. “Bayesian subspace methods for acoustic signature recognition of
vehicles”. In 12th European Signal Processing Conf), 2004
[53] H.-l. Wang, W. Yang, W.-d. Zhang, and Y. Jun. “Feature extraction of acoustic signal
based on wavelet analysis”. In ICESSSYMPOSIA ’08: Proceedings of the 2008
International Conference on Embedded Software and Systems Symposia. Washington,
DC, USA: IEEE Computer Society, 2008
[54] X. Shao, L. Sun. “An application of the continuous wavelet transform to resolution of
multicomponent overlapping analytical signals”. Analytical Letters, 34(2):267–280, 2001
[Online]. Available at: http://dx.doi.org/10.1081/AL-100001578
[55] H. C. Choe. “Signature detection, recognition, and classification using wavelets in
signal and image processing”. Ph.D. dissertation, Texas A&M University, Department of
Electrical Engineering, 1997
[56] R. Karlsen, T. Meitzler, G. Gerhart, D. Gorsich, and H. Choe. “Comparative study of
wavelet methods for ground vehicle signature analysis”. In Proceedings of the SPIE - The
International Society for Optical Engineering, 2762: 314–24, 1996
[57] Wang, Lipo, Fu, and Xiuju. “Data Mining with Computational Intelligence”. Springer
Berlin Heidelberg, 2005
[58] M. Wlchli and T. Braun. “Event classification and filtering of false alarms in wireless
sensor networks”. Parallel and Distributed Processing with Applications, International
Symposium on, 0:757–764, 2008
[59] D. Janakiram, V. A. M. Reddy and A. P. Kumar. “Outlier detection in wireless sensor
networks using bayesian belief networks”. In Communication System Software and

Middleware, First International Conference, 2006
[60] D. S. Kim, M. A. Azim, and J. S. Park. “Privacy preserving support vector machines in
wireless sensor networks”. In ARES ’08: Proceedings of the 2008 Third International
Conference on Availability, Reliability and Security Washington, DC, USA: IEEE Computer
Society, 2008
[61] X. Wang, D.-w. Bi, L. Ding, and S. Wang. “Agent collaborative target localization and
classification in wireless sensor networks. ”Sensors, 7(8):1359–1386, 2007. [Online].
Available at: http://www.mdpi.com/1424-8220/7/8/1359
[62] D. Tran and T. Nguyen. “Support vector classification strategies for localization in
sensor networks”. In Communications and Electronics, 2006. ICCE ’06. First International
Conference, 2006
[63] L. Yip, K. Comanor, J. C. Chen, R. E. Hudson, K. Yao, and L. Vandenberghe. “Array
processing for target doa, localization, and classification based on aml and svm algorithms
in sensor networks”. In AML and SVM Algorithms in Sensor Networks, Proc. of the
2nd International Workshop on Information Processing in Sensor Networks (IPSN03, 2003
[64] H. Wu and J. M. Mendel. “Classification of battlefield ground vehicles using acoustic
features and fuzzy logic rule-based classifiers”. Fuzzy Systems, IEEE Transactions 15(1):56
–72, 2007
[65] S. Thoerdoridis and K. Koutroumbas. “Pattern Recognition”. Elsvier Inc., 2009
[66] B. Lantow. “Applying distributed classification algorithms to wireless sensor networks
a brief view into the application of the sprint algorithm family”. In Networking, ICN, Seventh
[67] C. Meesookho, S. Narayanan, and C. Raghavendra. “Collaborative classification
applications in sensor networks”. In Sensor Array and Multichannel Signal Processing
Workshop Proceedings, 2002
[68] A. Prieto, C. G. Puntonet, B. Prieto, and M. Rodr´ıguez-A´ lvarez. “A competitive neural
network for blind separation of sources based on geometric properties”. In IWANN ’97:
Proceedings of the International Work-Conference on Artificial and Natural Neural
Networks. London, UK: Springer-Verlag, 1997
[69] Y.-W. Chen, X.-Y. Zeng, and Z. Nakao. “Blind separation based on an evolutionary
neural network”. Pattern Recognition, International Conference on, (2): p. 2973, 2000
[70] M. Solazzi and A. Uncini. “Spline neural networks for blind separation of post-
nonlinear-linear mixtures”. Circuits and Systems I: Regular Papers, IEEE Transactions on,
51(4):817–829, 2004
[71] J.-M. Ye, X.-L. Zhu, and X.-D. Zhang. “Adaptive blind separation with an unknown
number of sources”. Neural Comput., 16(8):1641–1660, 2004
[72] T.-Y. Sun, C.-C. Liu, S.-J. Tsai, and S.-T. Hsieh. “Blind source separation with dynamic
source number using adaptive neural algorithm”. Expert Syst. Appl., 36(5)8855–8861, 2009
[73] W. Penny, S. Roberts, and R. Everson. “Hidden markov independent components for
biosignal analysis”. In Advances in Medical Signal and Information Processing, 2000. First
International Conference on (IEE Conf. Publ. No. 476), 2000

[74] X. Wang, H. Qi, and H. Du. “Distributed source number estimation for multiple target
detection in sensor networks”. In Statistical Signal Processing, 2003 IEEE Workshop, 2003
[75] Y. Kai, H. Qi, W. Jianming, and L. Haitao. “Multiple vehicle signals separation
based on particle filtering in wireless sensor network”. Journal of Systems
Engineering and Electronics, 19(3):440-446, 2008. [Online]. Available at
:http://www.sciencedirect.com/science/article/B82XK-4SVTN2V-
5/2/379989fc284a479b6102967e30b0769a
[76] H. Chen, C. K. Tse, and J. Feng “Source extraction in bandwidth constrained wireless
sensor networks”. IEEETRANSACTION ON CIRCUITS AND SYSTEMS-II:EXPRESS
BRIEFS, 55(9):947–951, 2008
[77] H. Qi, X. Tao, and L. H. Tao. “Multiple target recognition based on blind source
separation and missing feature theory”. In Computational Advances in Multi-Sensor
Adaptive Processing 1st IEEE International Workshop on Volume, 2005
[78] F. Silva, J. Heidemann, R. Govindan, and D. Estrin. “Frontiers in Distributed Sensor
Networks”. CRC Press, Inc., 2003
[79] X. Wang, H. Qi, and H. Du. “Distributed source number estimation for multiple target
detection in sensor networks”. In StatisticalSignal Processing, 2003 IEEE Workshop, 2003
[80] R. J. Weiss and D. P. W. Ellis. “Estimating single-channel source separation masks:
Relevance vector machine classifiers vs. pitch-based masking”. In Proceedings of the
ISCA Tutorial and Research Workshop on Statistical and Perceptual Audition (SAPA), 2006
[81] R. Braunling, R. M. Jensen, M. A. Gallo. “Acoustic target detection, tracking,
classification, and location in a multiple-target environment”. G. Yonas, Ed., SPIE,
3081(1): 57–66, 1997 [Online]. Available at: http://link.aip.org/link/?PSI/3081/57/1
[82] E. Drakopoulos, J. J. Chao, C. C. Lee. “A two-level distributed multiple hypothesis
decision system”. 37(3):380–384,1992
[83] J. H. Kotecha, V. Ramachandranand, A. M. Sayeed. “Distributed multitarget
classification in wireless sensor networks”. 23(4):703–824, 2005
[84] E. F. Nakamura, A. A. F. Loureiro, A. C. Frery. “Information fusion for wireless
sensor networks: Methods, models, and classifications”. ACM Comput. Surv., 39(3):9,
2007
[85] R. Brooks, P. Ramanathan, A. Sayeed. “Distributed target classification and tracking in
sensor networks”. Proceedings of the IEEE, 91(8):1163–1171, 2003
.
[86] A. Aljaafreh and L. Dong. “Hidden markov model based classification approach for
multiple dynamic vehicles in wireless sensor networks”. In Networking, Sensing and
Control (ICNSC), 2010 International Conference, 2010
[87] J. Llinas and D. Hall. “An introduction to multi-sensor data fusion”. In Circuits and
Systems, ISCAS ’98. Proceedings of the 1998 IEEE International Symposium on 6:1998
[88] I. Liggins, M.E., C.-Y. Chong, I. Kadar, M. Alford, V. Vannicola, S. Thomopoulos.
“Distributed fusion architectures and algorithms for target tracking”. Proceedings of the
IEEE, 85(1):95–107, 997

[89] Z. H. Kamal, M. A. Salahuddin, A. K. Gupta, M. Terwilliger, V. Bhuse, and B.
Beckmann. “Analytical analysis of data and decision fusion in sensor networks”. In
ESA/VLSI, 2004
[90] G. Xing, R. Tan, B. Liu, J. Wang, X. Jia, and C.-W. Yi. “Data fusion improves the
coverage of wireless sensor networks”. In MobiCom ’09: Proceedings of the 15th annual
international conference on Mobile computing and networking. New York, NY, USA: ACM,
2009
[91] D. J. Miller, Y. Zhang, G. Kesidis. “Decision aggregation in distributed classification by
a transductive extension of maximum entropy/improved iterative scaling.” EURASIP
Journal on Advances in Signal Processing, 21: 2008
[92] A. D’Costa, V. Ramachandran, A. Sayeed. “Distributed classification of gaussian
space-time sources in wireless sensor networks”. Selected Areas in Communications, IEEE
Journal on, 22(6):1026–1036, 2004
[93] A. Aljaafreh and L. Dong. “Cooperative detection of moving targets in wireless sensor
network based on fuzzy dynamic weighted majority voting decision fusion”. In Networking,
Sensing and Control (ICNSC), International Conference, 2010
[94] J. Kittler, M. Hatef, R. P. Duin, and J. Matas. “On combining classifiers”. IEEE
Transactions on Pattern Analysis and Machine Intelligence, 20(3): 226–239, 1998
[95] X. Wang and S. Wang. “Collaborative signal processing for target tracking in distributed
wireless sensor networks”. J. Parallel Distrib. Comput., 67(5):501–515, 2007
[96] D. Hall, J. Llinas. “An introduction to multisensor data fusion”. Proceedings of the IEEE,
85(1)6–23,1997
[97] A. Sinha, H. Chen, D. Danu, T. Kirubarajan, and M. Farooq. “Estimation and decision
fusion: A survey.” Neurocomputing, 71(13-15):2650– 2656, 2008, artificial Neural Networks
(ICANN 2006) / Engineering of Intelligent Systems (ICEIS 2006). [Online]. Available at:
http://www.sciencedirect.com/science/article/B6V10-4SFS0KH-
8/2/f59b9b186c4bbbfe157bf08a07f72c4f.
[98] E. Drakopoulos, J. Chao, and C. Lee. “A two-level distributed multiple
hypothesis decision system”. Automatic Control, IEEE Transactions on, 37(3):380–
384,1992
[99] A. Speranzon, C. Fischione, and K. Johansson. “Distributed and collaborative
estimation over wireless sensor networks”. In Decision and Control, 45th IEEE Conference,
2006

Multi-Target Classification Using Acoustic Signatures in Wireless Sensor Networks: A survey

More Related Content

What's hot (19)

Similar to Multi-Target Classification Using Acoustic Signatures in Wireless Sensor Networks: A survey (20)

Recently uploaded (20)

Multi-Target Classification Using Acoustic Signatures in Wireless Sensor Networks: A survey