Multi-Channel Signal Segmentation and Classification

MULTI-CHANNEL SIGNAL SEGMENTATION AND CLASSIFICATION
Ales Prochazka and Veronika Bartosova

Department of Computing and Control Engineering
Institute of Chemical Technology in Prague
Technicka 6, 166 28 Prague 6, Czech Republic
Email: A.Prochazka@ieee.org
ABSTRACT
Multi-channel sensors and multi-channel signal analysis
form a specific area of general digital signal processing
methods with applications in medicine, environmental signal analysis or technology. The paper is devoted to general
mathematical methods related to initial signal de-noising,
detection of its principal components and segmentation to
find its specific parts. Feature detection includes the use
of discrete wavelet transform (DWT) and discrete Fourier
transform (DFT) for estimation of features invariant to signal shift to form clusters of close data segments. The selforganizing neural networks are then used for signal segments classification. Results are numerically evaluated by
statistical analysis of distances of individual feature vector
values from the corresponding cluster centers. Proposed
methods are used for electroencephalogram (EEG) signal
segmentation based upon detection of changes of signal
spectral components applied to its first principal component, signal segments feature extraction and their classification. Results achieved are compared for different data sets
and different mathematical methods used to detect signal
segments features. Numerical results are compared with
experience of experts specialized to EEG data analysis to
allow further correlation with MR images. Proposed methods are accompanied by the appropriate graphical user interface (GUI) designed in the MATLAB environment.
KEY WORDS
Multi-channel signal proceesing, discrete wavelet transform, principal component analysis, feature extraction,
biomedical signal and image analysis
Oldrich Vysata
Neurocenter Caregroup
Jiraskova 1389, 516 01 Rychnov nad Knez nou
Czech Republic
Email: vysata@neurol.cz
traction. The task of signal segments classification is
solved by neural networks [8, 5] in many cases. The
paper presents wavelet signal features classification by
self-organizing neural networks and it presents appropriate
graphical user interface proposed.
The information content of EEG signals [13, 12] is
fundamental for brain analysis and in connection with magnetic resonance methods it forms one of the most complex
diagnostic tools. To analyse extensive EEG observations
it is necessary to use efficient mathematical tools for fast
enough data processing. Digital filtering using finite impulse response (FIR) filters or discrete wavelet transform
can be used in the EEG signal preprocessing stage to remove power frequency from the observed signal and to reduce its undesirable frequency components at first. Fig. 1
presents a sample of an observed EEG signal segment with
results of its filtering in Fig. 2.
Signal segmentation methods related to change-point
detection can are usually applied to a given time series using Bayesian method [6] detecting changes of its statistical properties or it is possible to detect changes in signal frequency components important for EEG signal processing. Owing to the multi-channel basis of EEG signals
this method has been applied after the principal component
analysis (PCA) of an observed and preprocessed multichannel signal to its first principal component.
EEG SIGNAL: Data/18 Total Length: 847 s
DENOISED EEG SIGNAL Time Range: <25, 35> s
Figure 1. Original EEG signal recording presenting 19

channels with the additive
noise of 50 Hz
Figure 2. De-noised EEG

signal after the removal of
the signal noise components
INTRODUCTION
Multi-sensor processing of noisy non-stationary observations [16] form an important area of digital signal processing. Wavelet functions including dual-tree complex
wavelet transform [14, 1, 4] form an efficient mathematical tool for processing of such signals. While the use
of wavelet functions for multi-dimensional signals is described in many papers already their use for multi-channel
signals is still not so extensively studied.
The paper presents the use of wavelet functions in
combination with the principal component analysis [2, 15]
allowing their application for signal segments feature ex-
PRINCIPAL COMPONENT ANALYSIS
Principal component analysis [15] resulting from applied

linear algebra allows to reduce the dimension of a matrix
XN,M containing multi-channel data. In the case of EEG
records each its column j = 1, 2, , M represent channel
index and its row i = 1, 2, , N stand for observations.
It is possible to transform this matrix into a new one using
matrix PM,M to find values
YN,M = XN,M PM,M
(1)
In the case of principal component analysis matrix PM,M

is orthonormal evaluating columns of matrix YN,M with
the decreasing variance. Assuming that signal-to-noise
(SNR) ratio is defined by relation
SIGNAL SEGMENTATION
The first principal component of the multi-channel signal

can be used for segmentation of the whole set of observed
time-series. While Bayesian methods can be used for this
purpose [6] it is more efficient in the case of EEG signals to
detect changes of spectral components in the given signal.
The proposed method uses two sliding overlapping windows and the whole algorithm is based upon detection of
changes of spectral components in given frequency bands.
Fig. 5 presents the proposed graphical user interface designed to find signal segments of similar properties in the
frequency domain.
2
signal
(2)
2
noise
it is possible to assume that this analysis finds signal components with decreasing SNR. Proposed algorithm presented in Fig. 3 includes multi-channel signal filtering to
reject signal component of 50 Hz and to preserve signal
frequencies in the range of 0.5, 60 Hz and the following
evaluation of principal components.
SN R =
%% EEG Data Filtering %%%

b1=fir1(M,[48 52]/Fs,stop);
b2=fir1(M,[0.5 60]/Fs,bandpass);
X=filter(b2,1,filter(b1,1,EEG));
%% Principal Component Analysis %%%
P = princomp(X);
Y=X*P;
Figure 3. Algorithm of multi-channel FIR filtering (of order M ) of an EEG signal observed with the sampling frequency Fs and its following principal component analysis
Results of PCA for a selected segment of EEG observations
after their de-noising is presented in Fig. 4 together with
resulting signal variances.
Figure 4. Graphical user interface for principal component

analysis of the set of observed EEG channels and evaluation of results
Figure 5. Graphical user interface for segmentation of the

EEG signal and average frequency energy of the selected
segment in given frequency bands
WAVELET FEATURE EXTRACTION
Discrete wavelet transform (DWT) forms a general mathematical tool for signal processing with many applications
in EEG data analysis [7, 9, 14, 3] as well. Its basic use includes time-scale signal analysis, signal decomposition and
signal compression.
The set of wavelet functions is usually derived from
the initial (mother) wavelet h(t) which is dilated by value
a = 2m , translated by constant b = k 2m and normalized
so that
1
tb
1
)=
hm,k (t) = h(
h(2m t k) (3)
a
a
2m
for integer values of m, k and the initial wavelet defined
either by the solution of a dilation equation or by an analytical expression [11]. The initial wavelet can be considered
as a pass-band filter and in most cases half-band filter covering the normalized frequency band 0.25, 0.5). A wavelet
dilation by the factor a = 2m corresponds to a pass-band
compression.
The set of wavelets define a special filter bank which
can be used for signal component analysis and resulting
wavelet transform coefficients can be further applied as signal features for its classification. Signal decomposition per-
formed by a pyramidal algorithm is interpreting wavelets

as pass-band filters. Another approach [11] is based upon a
very efficient parallel algorithm using the fast Fourier transform.
The proposed algorithm is based upon the wavelet decomposition of the signal segments and evaluation of its coefficients for their classification. Fig. 6 presents application
of this method to EEG signal segments and their analysis
by a harmonic wavelet transform [11] resulting in features
standing for scales 1, 2 and 3 respectively covering three
frequency bands with different time-scale resolutions. Another approach proposed [14] is based upon the dual tree
complex wavelet transform (DTCWT) use to reduce poor
shift sensitivity of the original DWT. The basic idea of this
method is based upon the signal decomposition running
in two parallel trees using real filters to generate real and
imaginary parts of complex coefficients.
Results of signal classification into four classes by a

self-organizing neural network are given in Fig. 7 for two
selected signal features allowing a simple visualization of
segmentation results and visualization of typical class representatives with their features closest to the corresponding
cluster centers.
(a) SEGMENTED SIGNAL
1
0.5
0
0.5
1
Index
10
12
Figure 7. Graphical user interface for signal segments classification into the given number of classes and detection of
typical signal segments closest to cluster centers
14
Segment
(b) SIGNAL FEATURES

Frequency
0.5
0.4
0.3
0.2
0.1
2
10
Segment
12
14
16
Figure 6. Results of feature extraction presenting (a) EEG

signal segments and (b) their wavelet features resulting
from a harmonic DWT on scales 1, 2, 3
Numerical results of classification of Q signal segments with feature matrix PR,Q = [p1 , p2 , , pQ ] for
the selection of different sets of R = 2 features and C = 4
classes are compared in Tab. 1. Each class i = 1, 2, , C
can be characterized by the mean distance of column feature vectors pjk belonging to class segments jk for k =
1, 2, , Ni from the class centre in the ith row of matrix WC,R = [w1 , w2 , , wC ] by relation
ClassDist(i) =
Ni
1
dist(pjk , wi )
Ni
(4)
k=1
CLASSIFICATION
Classification of signal segments into a given number of

classes using segments features can be achieved by various
statistical methods. Another approach is based upon the
application of self-organizing neural networks using features as patterns for the input layer of neural network. The
number of output layer elements is equal to signal classes
and must be either defined in advance or it can be automatically increased to create new classes [5]. During the learning process neural network weights are changed to minimize distances between each input vector and corresponding weights of a winning neuron characterized by its coefficients closest to the current pattern. In case that the learning process is successfully completed network weights belonging to separate output elements represent typical class
individuals.
where Ni represents the number of segments belonging to

class i and function dist is used for evaluation of the Euclidean distance between two vectors. Results of classification can be numerically characterized by the mean value of
average class distances related to the mean value of class
centers distances obtained after the learning process according to relation
crit = mean(ClassDist)/mean(dist(W, W ))
(5)
This proposed Cluster Segmentation Criterion (CSC) provides low values for compact and well separated clusters
while close clusters with extensive dispersion of cluster
vectors provide high values of this criterion. It is obvious
that classification parameters achieved and summarized in
Tab. 1 both by the DFT and DWT provide similar results
but slightly better in the case of wavelet features selection.
Table 1. Comparison of signal segments classification into

four classes using two features resulting from a chosen signal segments analysis
Features
Cluster Segmentation Criterion

Set 1
Set 2
Set 3
Set 4
(Q=21)
(Q=36)
(Q=71)
(Q=99)
DFT
0.31
0.61
0.53
0.55
DWT/db4
0.32
0.37
0.35
0.47
DWT/harmonic
0.21
0.22
0.25
0.23
CONCLUSION
The paper presents selected aspects of multi-channel signal

processing including the application of principal component analysis to define time-series used for signal changepoints detection. Mathematical methods used in this connection include discrete wavelet transform as a tool for signal de-noising and signal segments feature extraction to
define the pattern matrix for feature vectors classification.
Dual-tree complex wavelet transform is mentioned as well
as a tool for shift-invariant features estimation.
Methods proposed are applied to EEG signal denoising, their segmentation using principal component
analysis and classification of feature vectors. Signal classification proposed assumes the knowledge of the number of
classes while self-creating neural network structures used
for classification [5] are able to find the optimal value of
classes and to exclude the possibility of dead neurons.
The paper studies the influence of signal features
definition to the compactness of resulting clusters using
wavelet transform as a tool for signal segments analysis
in the time-scale domain. Proposed methods of multichannel signal analysis including their filtering, principal
component analysis and signal features classification are
designed as graphical user interfaces in the MATLAB environment to allow simple signal analysis and processing.
The compactness of resulting clusters is numerically studied through distances of class individuals and their center.
Further research will be devoted to the more detail
analysis of signal segments features to find clusters with
their individuals as close to cluster centers as possible.
Further studies will be devoted to extensive application of
methods proposed including the study of correlation between EEG observations and MR image analysis.
ACKNOWLEDGEMENT
Real EEG signals have been observed by the Neurocenter
Caregroup in Rychnov nad Knez nou. The dual-tree complex wavelet transform library has been kindly provided by
prof. N. Kingsbury [10] from the Engineering Department
of the University of Cambridge, U.K. The paper has been
supported by the Research grant No. MSM 6046137306.
References
[1] R. Anderson, N. Kingsbury, and J. Fauqueur.
Rotation-Invariant Object Recognition Using Edge
Profile Clusters. In The European Signal Processing
Conference EUSIPCO-06. EURASIP, 2006.
[2] C. M. Bishop. Neural Networks for Pattern Recognition. Clarendon Press, Oxford, U.K., 1995.
[3] N.P. Castellanos and V.A. Makarov. Recovering EEG
Brain Signals: Artifact Suppression with Wavelet Enhanced Independent Component Analysis. Journal of
Neuroscience Methods, 158(2):300312, 2006.
[4] C. Chaux, J. C. Pesquet, and L. Duval. Noise Covariance Properties in Dual-Tree Wavelet Decomposition. IEEE Transaction on Information Theory,
53(12):46904700, December 2007.
[5] D. I. Choi and S. H. Park. Self-Creating and Organizing Neural Networks. IEEE Trans. Neural Networks,
5(4):561575, July 1994.
[6] W. J. Fitzgerald, J. J. K. O. Ruanaidh, and J. A. Yates.
Generalised Changepoint Detection. Technical report, University of Cambridge, U.K., 1994.
[7] A. Glavinovitch, M.N.S. Swamy, and E.I. Plotkin.
Wavelet-Based Segmentation Techniques in the Detection of Microarousals in the Sleep EEG. In 48th
Midwest Symposium on Circuits and Systems, pages
13021305. IEEE, 2005.
[8] S. Haykin. Neural Networks. IEEE Press, NY, 1994.
[9] P. Johankhani, V. Kodogiannis, and K. Revett. EEG
Signal Classification Using Wavelet Feature Extraction and Neural Networks. In Int. Symp. on Modern
Computing, JVA06, pages 120124. IEEE, 2006.
[10] N. G. Kingsbury. Complex Wavelets for Shift Invariant Analysis and Filtering of Signals. Journal
of Applied and Computational Harmonic Analysis,
10(3):234253, May 2001.
[11] D. E. Newland. An Introduction to Random Vibrations, Spectral and Wavelet Analysis. Longman Scientific & Technical, Essex, U.K., third edition, 1994.
[12] M. Nixon and A. Aguado. Feature Extraction & Image Processing. Elsevier, Amsterdam, 2004.
[13] Saeid Sanei and J. A. Chambers. EEG Signal Processing. Wiley - Interscience, 2007.
[14] I. W. Selesnick, R. G. Baraniuk, and N. G. Kingsbury.
The Dual-Tree Complex Wavelet Transform. IEEE
Signal Processing Magazine, 22(6):123151, 2005.
[15] J. Shlens.
A Tutorial on Principal Component
Analysis. http://www.snl.salk.edu/ shlens/pub/notes/
pca.pdf, 2005.
[16] S. V. Vaseghi. Advanced Digital Signal Processing
and Noise Reduction. John Wiley & Sons Ltd, 2006.

Multi-Channel Signal Segmentation and Classification

Uploaded by

Copyright:

Available Formats

Multi-Channel Signal Segmentation and Classification

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multi-Channel Signal Segmentation and Classification

Uploaded by

Copyright:

Available Formats

MULTI-CHANNEL SIGNAL SEGMENTATION AND CLASSIFICATION

Ales Prochazka and Veronika Bartosova

EEG SIGNAL: Data/18 Total Length: 847 s

DENOISED EEG SIGNAL Time Range: <25, 35> s

Figure 1. Original EEG signal recording presenting 19

Figure 2. De-noised EEG

PRINCIPAL COMPONENT ANALYSIS

Principal component analysis [15] resulting from applied

In the case of principal component analysis matrix PM,M

The first principal component of the multi-channel signal

%% EEG Data Filtering %%%

Figure 4. Graphical user interface for principal component

Figure 5. Graphical user interface for segmentation of the

WAVELET FEATURE EXTRACTION

formed by a pyramidal algorithm is interpreting wavelets

Results of signal classification into four classes by a

(a) SEGMENTED SIGNAL

(b) SIGNAL FEATURES

Figure 6. Results of feature extraction presenting (a) EEG

Classification of signal segments into a given number of

where Ni represents the number of segments belonging to

Table 1. Comparison of signal segments classification into

Cluster Segmentation Criterion

The paper presents selected aspects of multi-channel signal

You might also like