ElectroencephalographySignalClassification based on Sub-Band Common Spatial Pattern (SBCSP)

IOSR Journal of VLSI and Signal Processing (IOSR-JVSP)
Volume 7, Issue 2, Ver. I (Mar. - Apr. 2017), PP 74-82
e-ISSN: 2319 – 4200, p-ISSN No. : 2319 – 4197
www.iosrjournals.org
DOI: 10.9790/4200-0702017482 www.iosrjournals.org 74 | Page
ElectroencephalographySignalClassification based on Sub-Band
Common Spatial Pattern (SBCSP)
Md. Sujan Ali1
, Mst. Jannatul Ferdous1
1
Department of Computer Science and Engineering, JatiyaKabiKaziNazrul Islam University, Mymensingh,
Bangladesh
Abstract:Brain-computer interface (BCI) is a communication pathway between brain and an external device. It
translates human thought into commands to control the external devices.Electroencephalography (EEG) is cost
effective and easier way to implement the BCI. This paper presents a novel method for classifying EEG during
motor imagery by the combination of common spatial pattern (CSP) and linear discriminant analysis (LDA). In
the proposed method, the EEG signal is bandpass-filtered into multiple frequency bands. The CSP features are
then extracted from each of these bands. The LDA classifier is subsequently used to classify the CSP features. In
this paper, experimental results are presented on a publicly available BCI competition dataset and the
performance is compared with existing approaches. The experimental result shows that the proposed method
yields comparatively superior cross validation accuracies compared to prevailing methods.
Keywords:brain computer interface, electroencephalography, sub-band common spatial pattern.
I. Introduction
Brain-computer interface (BCI) is a communicating system between a brain and a device that enables
signals from the brain to direct some external devices, such as a computer, wheelchairs [1], robotic arms,
prostheses [2] etc. The interface translates human thoughts into command to control the external
devices.Thekeytarget of the BCI is to restore or repair useful function to people disabled by neuromuscular
disorders such as Amyotrophic Lateral Sclerosis (ALS), cerebral palsy, stroke, or spinal cord injury.Although
people may become totally paralyzed through these types of disorders their minds are un-affected. Considering
this issue brain computer interface translates human thoughts directly to the external world [3]. An
electroencephalography (EEG) is the recorded electrical activity generated in the brain which is recorded by the
electrodes placing on the scalp.
Common Spatial Patterns (CSP) is an algorithm commonly used in BCI systems to preprocess the
electroencephalogram (EEG) signals [4, 5, 6].The algorithm finds optimal spatial filters that are functional in
discriminating two classes of EEG signals in motor imagery based BCI. The effectiveness of the spatial filters
depends on its subject specificfrequency band.If the EEG signalsis unfiltered or have been filtered with badly
chosen frequency rangethen the classification of that signals using CSP shows poor accuracies [7].
Consequently, subjectspecific frequency bands are generally used with the CSP algorithm [8].
To overcome the limitation of manually selecting the subject specific frequency bands for the CSP, the
Common Spatio-Spectral Pattern (CSSP) algorithm has been proposed where simple filters are optimized
together with the CSP algorithm [9]. The Common Sparse Spectral Spatial Pattern (CSSSP) algorithm improves
the performance of CSSP algorithm. It allows concurrent optimization of an arbitrary Finite Impulse Response
(FIR) filter within CSP analysis [8]. Another approach called SPEC-trally weighted Common Spatial Pattern
(SPEC-CSP) algorithm [10] optimizes the temporal filter in the frequency domain and after that the spatial filter
in an iterative method [11]. However, due to the inherent nature of optimization problem, the solution of filter
coefficients depends significantly on the selection of initial parameters [7].
Sub-band Common Spatial Pattern (SBCSP) method [7] was alternatively proposed and has been
shown better classification accuracy compared against CSSP and CSSSP. In this method publicly available
dataset from BCI competition III in 2005 has been used. As a substitute of temporal FIR filter within the CSP
algorithm, SBCSP uses a filter bank that decomposes the EEG signals into sub-bands. The CSP algorithm is
then employed on each of these sub-bands to obtain sub-band scores. To fuse the sub-band score two fusion
methods namely Recursive Band Elimination (RBE) and Meta-Classifier (MC) are used. An additional classifier
is then usedto classify the fused sub-band scores. In [7] comparative study of using different sub-band score
fusion techniques and classification algorithms are not available [12].
The Filter BankCommon Spatial Pattern (FBCSP) algorithm [12] wasproposed to classify EEG signals.
In the algorithm, the EEG signals are bandpass-filtered into some frequency bands and CSP features are
extracted from each of these bands. Finally, a classification algorithm is used to classify the selected CSP
features.The FBCSP algorithm used the typical estimation ofmultivariate covariance matrices from the

Electroencephalography Signal Classification based on Sub-Band Common Spatial Pattern (SBCSP)
EEGsignals for a filter bank of CSP [12]. Usually EEG signals are contaminated with artifacts or different types
of noise sources. Due to the contamination the normal pattern of the majority of the EEG data are differed [13].
In the case of large amount of contamination, the multivariate covariance estimates typically varies significantly
from the estimate without the contamination [13]. Therefore, the FBCSP algorithm is sensitive to artifacts in the
trainingdata [14].
A Robust Filter Bank CommonSpatial Pattern (RFBCSP) algorithm was proposed [14] where
theMinimum Covariance Determinant (MCD) estimator is used to estimate the covariance matrices. Likewise,
to estimate the variance of the projected EEG signals the Median Absolute Deviation (MAD) is used. The
classification performance of the RFBCSP is better in some specific subjects but the overall results are not
statistically significant.
In this paper, a novel approach is proposed for EEG signal classification in motor imagery-based BCI.
The proposed approach is subdivided into the following three stages. In the first stage, the EEG signal is divided
into multiple frequency bands using bandpass filter. In the second stage, CSP features are extracted from each of
these frequency bands. A classification algorithm is used to classify the CSP features in the third stage. In the
third stage, the classification of each band is done by three steps: finding Linear Discriminant Analysis (LDA)
scores, blending LDA scores and classifying based on the LDA scores.
The paper is organized as follows– Section II discusses a feature extraction technique called Common
Spatial Pattern (CSP), the Linear Discriminant Analysis (LDA) technique is explained in section III, section IV
contains the description of the proposed method, the experimental results are illustrated in section V and the
section VI includes some concluding remarks.
II. Common Spatial Pattern
Common Spatial Pattern (CSP) is a feature extraction technique used in signal processing for
separating a multivariate signal into additive subcomponents.The technique used to design spatial filters such
that the variance of the filtered data from one class is maximized while the variance of the filtered data from the
other class is minimized. Thus, the resulting feature vectors increase the discriminability between the two
classes by means of minimize the intra class variance and maximize the inter class variance [15]. This property
builds CSP as one of the most effective spatial filters for EEG signal processing. The method of CSP was first
introduced to EEG analysis for detection of abnormal EEG [16] and effectively applied on movement-related
EEG for the classification purpose [4, 6]. The target of the CSP is to project the multichannel EEG data into low
dimensional spatial subspace with a projection matrix using linear transformation [17].
For details explanation of the CSP algorithm, assume the original EEG data matrix i
kE from trial i for
class k. The dimension of each i
kE is TN  , where N is the number of channels and T is the number of samples
per channel. For the explanation, the EEG data of a single trial )1( i is represented as ),( rlkE  where l denotes
left hand and r denotes right hand movement. The normalized spatial covariance of the EEG for the left hand
movement, lC and for the right hand movement, rC can be calculated as:
l
T
ll
l
S
EE
C  ,
r
T
rr
r
S
EE
C  (1)
Where lE and rE represent the original EEG matrices for left hand and right hand movement respectively, T
lE
is the transpose of lE and T
rE is the transpose of rE .The  T
lll EEtraceS  and  T
rrr EEtraceS  are the
sum of the diagonal elements of T
ll EE and T
rr EE respectively. The composite spatial covariance, C is the sum
of the averaged normalized spatial covariance lC and rC . The lC and rC are estimated by averaging over all the
trials of each class. The composite spatial covariance, C is calculated as
T
eeerl MMCCC  (2)
Where eM is the matrix of eigenvectors, T
eM is the transpose of eM and e is the diagonal matrix of
eigenvalues.
The averaged normalized spatial covariance lC and rC are transformed as
T
ll XCXJ  and T
rr XCXJ  (3)

Where e
T
eMX / is the whitening transformation matrix and its transpose is T
X . lJ and rJ share common
eigenvectors and the sum of corresponding eigenvalues for the two matrices will always be one. If T
ll YYJ 
and T
rr YYJ  then Irl  , where I is the identity matrix. Since the sum of two corresponding
eigenvalues is always one, a high eigenvalue for lJ means that a high variance for EEG in left hand movement
and a low variance for the EEG in right hand movement (low eigenvalue for rJ ) and vice versa. The
classification operation is done based on this property. The projection of whitened EEG onto the eigenvectors Y
corresponding to the largest l and r will give feature vectors that significantly enhance the discrimination
ability.
The goal of the CSP is to find F spatial filters to create a projection matrix W of dimension FN  (each column
is a spatial filter). The projection matrix W is represented as
XYW T
 (4)
The projection matrix W linearly transforms the original EEG into uncorrelated components according to:
WEZ  (5)
The original EEG, E can be reconstructed by ZWE 1
 where 1
W is the inverse matrix of W. The columns of
1
W are spatial patterns that describe the variance of the EEG. The first and last columns contain the most
discriminatory spatial patterns that explain the high variance of one class and the low variance of the other.
III. Linear Discriminant Analysis
Linear Discriminant Analysis (LDA), also known as Fisher‟s linear discriminant analysis is a technique
used to find a linear combination of features that separates two or more classes of data. It is typically used as a
dimensionality reduction step before classification [18]. It reduces dimensionality but at the same time preserves
as much of the class discriminatory information as possible. The goal of the LDA is to use a
separatinghyperplane that maximally separate the data representing the different classes. The hyperplane is
found by selecting the projectionwhere the same classes are projected very close to each other and the distance
between the two classes means is as maximum as possible [19]. An example of a selection of data projection is
shown in Fig. 1. As shown in Fig. 1 projection p1 is a better line where class 1 and class 2 are well separated
whereas projection p2 line is unable to separate the two classes.
Figure 1: An example of a selection of data projection. Projection p1 maximize the separation of data compare
to projection p2
Let as assume that we have Kclasses, each containing N observations xi. The within-class scatter, wS
~
for all K classes can be calculated as:


K
k
k
wkw SfS
1
~
(6)
Where the within-class covariance matrix k
wS , the fraction of data kf and the mean vector k of class kare
calculated according to the following formulas:
Tkk
i
N
i
kk
i
k
w xxS
k
)()(
1
  

(7)




K
j
j
k
k
N
N
f
1
, 

kN
i
k
i
k
k x
N 1
1
 (8)
The between class scatter bS
~
for all K classes can be calculated as:


K
k
k
bkb SfS
1
~
(9)
Where the between class covariance matrix, k
bS for the mean of all observations xi for all K classes,  can be
estimated as
Tk
K
k
kk
bS )()(
1
  

(10)
The main objective of LDA is to find a projection matrix that maximizes the ratio of the determinantof bS
~
to
the determinant of wS
~
. The projections that providing the best class separation are eigenvectors with the highest
eigenvalues of matrixP [18]:
w
b
S
S
P ~
~
 (11)
Figure 2:Block diagram of the proposed EEG signal classification approach
Since the matrix P is asymmetric, the calculation of eigenvectors can be difficult. This difficulty can be
minimized by using generalized eigenvalue problem [20]. Now, the aim of the LDA is to seek (K-1) projections
 1321 ,...,,, Kyyyy by means of (K-1) projection vectors. The transformed data set y is obtained as a linear
combination of all input features x with weights W.
Wxy T
 (12)
Where  DwwwwW ,...,,, 321 is a matrix form with the D eigenvectors of matrix P associated with the
highest eigenvalues. The LDA reduces the original feature space dimension to D. The LDA performs well when
the discriminatory information of data depends on the mean of the data. But it does not work for the variance
depended discriminatory informative data. Also, the performance of the LDA is not good for nonlinear
classification.

IV. Proposed Approach
The proposed EEG signal classification approach is illustrated in Fig. 2.This approach is subdivided
into three stages for EEG signal processing and machine learning. In the first stage, the EEG signal frequency is
filtered into multiple pass bands using bandpass filter. In the second stage, CSP features are extracted from each
of these frequency bands. In the third stage, the classification operation is performed by finding LDA scores,
blending and classifying the scores. A detail of each stage is described in below.
Frequency filtering: the first stage filters the EEG signal into multiple frequency passbands. The digital
Butterworth bandpass filter is used to filter the EEG signal. Here, the most dominating rhythmic components
alpha and beta (8-32Hz) are selected. A total of sixbandpass filters 8-12Hz, 12-16Hz, 16-20Hz, 20-24Hz, 24-
28Hz and 28-32Hz are used. The filtered sixsubbands are used individually for the classification.
Spatial filtering:in this stage the CSP algorithm is used to perform the spatial filtering operation. The
spatial filter produces CSP features for the particular frequency range of each of the sub bands. Classification:In
the third stage, classification algorithms called LDA classifier is used to model and classify the selected CSP
features. Each sub band feature is passed separately through the classifier. To validate the classification, QP 
cross validation is used. At the first step of the classification stage, the LDA classifier computed LDA scores for
every value of P and Q. In the score mixture step, the LDA scores are mixed up according to



Q
iQ 1
1
(13)
Where,  and  denote the mixed LDA scores and the LDA scores computed by the LDA classifier
respectively. The mixed LDA scores are converted to predicted classes. The accuracy, Qj is tested based on
the predicted classes in the score classification step where Pj ,...,2,1 . After QP  cross validation, the
classification accuracy for each subbandis estimated by the following formula:



P
j
Qjb
P 1
1
(14)
Where, b is the classification accuracy for subband ),..,2,1( Bbb  . Finally, the classification rate (CR) of the
EEG signal is calculated according to (15)
%100)max(  bCR (15)
V. Experimental Results
The performance of the proposed method is evaluated by classifying EEG during imagined movement.
The proposed approach is applied to the publicly available BCI competition dataset. A filter bank is used in this
method that covers alpha and beta rhythmic components (8-32Hz). The filter bank comprises sixbandpass filters
namely 8-12Hz, 12-16Hz, 16-20Hz, 20-24Hz, 24-28Hz and 28-32Hz. A fourth-order Butterworth filter is used
to subband the EEG data.To extract features from the data, the CSP algorithm with 2m is used in this
experiment.
Dataset: To evaluate the performance of the proposed method, the dataset IVa from the publicly
available BCI competition III 2005 [21] is used in this experiment. This dataset contains data from the four
initial sessions without feedback. The dataset is recorded from five healthy subjects (labelled „aa‟, „al‟, „av‟,
„aw‟, „ay‟) who performed right hand and right foot movement imagination [22]. The data for each subject
comprises 280 trials from 118 EEG channels and 140 trials in each class. The visual cues at each trial last for 3.5
seconds. The sampling rate of the data is 100 Hz. In this experiment, the data between 0.5 seconds and 2.5
seconds from the visual cue (i.e. 200 time points at each trial) is extracted.
Channel selection: The motor imagery response of brain is more active in its central part [23]. In this
experiment, out of the 118 EEG channels, from the central area 13 are selected for classification. The selected
EEG channels are “FC3”, “FC4”, “Cz”, “C1”, “C2”, “C3”, “C4”, “C5”, “C6”, “T7”, “T8”, “CP3”, and “CP4”.
The spatial distribution of the channels on the scalp in 10/20 EEG system is illustrated in Fig. 3. The letters
specify the spatial location as A = ear lobe; C = central; P = parietal; F = frontal; T = temporal; O = occipital; FP
= frontal polar; FC = between frontal andcentral; CP = between central and parietal; FT = between frontal and
temporal; PO = between parietal and occipitaland AF= intermediate between frontal polar and frontal. The
channels used in this experiment are indicated by the circle of bold line in Fig. 3.

Figure 3: Location and nomenclature of the intermediate 10% electrodes (10/20 EEG system), as standardized
by the American EEG society.
Figure 4: The spectrum of (a-b) subband 1(8-12Hz) and (c-d) subband 2(12-16Hz) components for right hand and right
foot movement respectively.
Figure 5: The spectrum of (a-b) subband3(16-20Hz) and (c-d) subband4(20-24Hz) components for right hand and right

Figure 6: The spectrum of (a-b) subband5(24-28Hz) and (c-d) subband6(28-32Hz) components for right hand and right
Subject
Subband 1 (8-12Hz) Subband 2 (12-16Hz)
Right hand
movement
Right foot
movement
Right hand
movement
Right foot
movement
aa
al
av
aw
ay
Figure 7: Topographical map of brain for subband 1(8-12Hz) and subband
2(12-16Hz) of five subjects; first trace: right hand, second trace: right foot
movement of subband 1; third trace: right hand, fourth trace: right foot
movement of subband 2.

Table I: Classification accuracy (%)
Method subject
aa al av aw ay average
CSP1 71.3 88.4 48.6 89.9 79.9 75.6
CSP2 65.3 90.2 63.7 80.3 87.3 77.4
EMD-CSP 68.4 89.6 64.1 82.5 86.9 78.3
MEMD1-CSP 68.8 90.0 68.8 76.3 87.5 78.3
MEMD2-CSP 60.3 82.9 55.3 60.7 74.0 66.6
FBCSPw 93.3 98.5 66.8 93.8 93.6 89.2
FBCSPf 86.0 97.9 76.8 96.8 94.0 90.3
Proposed Method 94.4 98.7 81.2 98.2 96.8 93.9
In Fig. 4-6, the individual color line indicates the energy (normalized) contributed by the different
subbands to the selected channels. Fig. 4 (a) and Fig. 4 (b) show the spectrums of the subband 1 (8-12Hz)
component for right hand and right foot movement respectively. The spectrums of the subband 2 (12-16Hz) for
right hand and right foot movement are depicts in Fig. 4 (c) and Fig. 4 (d) respectively. Each color trace in Fig.
5 represents the spectrum of the activity of subband 3 (16-20 Hz) and subband 4 (20-24Hz) for both right hand
and right foot movement. The spectrums of the higher frequency subbands, subband 5 (24-28Hz) and subband 6
(28-32Hz) are shown in Fig. 6. From Fig. 4-6, the overall observation is that for right hand movement channels
C2, C3, Cz, C4, FC4 and T8 shows comparatively more energy than other channels. On the other hand, channels
C3, C4, CP3, PC3, FC4 and T7 shows comparatively more energy than rest of the channels for the right foot
movement.
The topographical brain maps for subband 1 and subband 2 during imaginary right hand and right foot
movement for the five subjects („aa‟, „al‟, „av‟, „aw‟, „ay‟) are shown in Fig. 7. The most significant CSP of the
two subbands are used for the topographical brain maps. The first and second trace (Fig. 7) shows the
topographical brain maps of the subband 1 for imaginary right hand and foot movement respectively. The
topographical brain maps of subband 2 for the imaginary right hand and foot are shown in third and fourth trace
(Fig. 7) respectively. From Fig. 7 we observed that for right hand movement the electrodes of right hemisphere
of the head scalp are more active whereas the electrodes of left hemisphere of the head scalp are more active for
the right foot movement.
Classification results:In this paper, we found the classification results of the EEG during imagined right
hand and right foot movement using the proposed method. Table I shows the classification accuracy of unbiased
10×10–fold cross validations performed. We compare the performance of the proposed method to that of the
other methods (CSP, EMD-CSP and MEMD-CSP) proposed in [24] and methods (FBCSPw, FBCSPf) proposed
in [12].Table Ishow that our proposed method yields superior result than all othermethods.
VI.Conclusion
A novel method to classify EEG during imagined right hand and right foot movement is introduced in
this paper. In this method the EEG is filtered into multiple sub ands for the purpose of selecting an appropriate
operational frequency band. The discriminative CSP features are then extracted from each of these subbands.To
classify the extracted features a classification algorithm, LDA is used. LDA score is produced for every fold
cross validation. The LDA scores are mixed up and the mixed scores are converted to predicted class. The
classification accuracy is tested based on the predicted class. The experimental results show that the proposed
method yields superior classification accuracy compared against existing methods CSP, EMD-CSP, MEMD-
CSP, FBCSPwandFBCSPf.
Acknowledgements
This research work supported by the Information and Communication Technology (ICT) division of the
ministry of Post, Telecommunication and Information Technology, Bangladesh.
References
[1]. B. Rebsamen, E. Burdet, C. Guan, H. Zhang, C. L. Teo, Q.Zeng, C.Laugier, and M. H. Ang Jr., "Controlling a Wheelchair Indoors
UsingThought," IEEE IntelligentSystems, vol. 22, no. 2, pp. 18-24, 2007.
[2]. N. Birbaumer, "Brain-computer-interface research: Coming of age," Clin. Neurophysiol., vol. 117, no. 3, pp. 479-483, 2006.
[3]. J. R. Wolpaw, N. Birbaumer, D. J. McFarland, G. Pfurtscheller and T. M. Vaughan, “Brain-computer interfaces for communication
andcontrol,”Clin. Neurophys., vol. 113, pp. 767-791, 2002.
[4]. H. Ramoser, J. M¨uller-Gerking, and G. Pfurtscheller, “Optimal spatial filtering of single trial EEG during imagined hand
movement,” IEEE Transactions on Rehabilitation Engineering, vol. 8, no. 4, pp. 441–447, 2000.
[5]. G. Dornhege, B. Blankertz, M. Krauledat, F. Losch, G. Curio, and K.- R. M¨uller, “Optimizing spatio-temporal filters for improving
Brain-Computer Interfacing,” in Advances in Neural Inf. Proc. Systems (NIPS05), J. Platt, Ed., vol. 18, Vancouver, Canada,
December 2005.

[6]. J. Müller-Gerking, G. Pfurtscheller, and H. Flyvbjerg, “Designing optimalspatial filters for single-trial EEG classification in a
movementtask,” Electroenc.Clin.Neurophys., pp. 787-798, 1999.
[7]. Q. Novi, C. Guan, T. H. Dat, and P. Xue, "Sub-band Common Spatial Pattern (SBCSP) for Brain-Computer Interface,"3rd
InternationalIEEE/EMBS Conference on Neural Engineering, 2007.CNE '07, pp. 204-207, 2007.
[8]. G. Dornhege, B. Blankertz, M. Krauledat, F. Losch, G. Curio, and K. R. Muller, "Combined Optimization of Spatial and Temporal
Filters for Improving Brain-Computer Interfacing," IEEE Trans. Biomed. Eng., vol. 53, no. 11, pp. 2274-2281, 2006.
[9]. S. Lemm, B. Blankertz, G. Curio, and K.-R. Muller, "Spatio-Spectral Filters for Improving the Classification of Single Trial EEG,"
IEEETrans. Biomed. Eng., vol. 52, no. 9, pp. 1541-1548, 2005.
[10]. R. Tomioka, G. Dornhege, G. Nolte, B. Blankertz, K. Aihara, and K.-R Müller, “Spectrally Weighted Common Spatial Pattern
Algorithm for SingleTrialEEGClassification”,Mathematical Engineering Technical Reports,UniversityofTokyo, Tokyo, 2006.
[11]. K. K. Ang, Z. Y. Chin, C. Wang, C. Guan and H. Zhang, “Filter bank common spatial pattern algorithm on BCI competition IV
Datasets 2a and 2b,” frontiers in NEUROSCIENCE, vol. 6, 2012.
[12]. K. K. Ang, Z. Y. Chin, H. Zhang, and C. Guan, "Filter Bank Common Spatial Pattern (FBCSP) in Brain-Computer Interface," in
Proc.IJCNN'08, pp. 2391-2398, 2008.
[13]. M. Hubert, P. J. Rousseeuw, and S. Van Aelst, "High-Breakdown Robust Multivariate Methods," Stat. Sci., vol. 23, no. 1, pp. 92-
119, 2008.
[14]. K. K. Ang, Z. Y. Chin, H. Zhang, and C. Guan, „”Robust Filter Bank Common Spatial Pattern (RFBCSP) in motor-imagery-based
Brain-Computer Interface”, 31st Annual International Conference of the IEEE EMBSMinneapolis, Minnesota, USA, pp. 578-581,
September 2-6, 2009.
[15]. Rajesh P. N. Rao, “Brain-Computer Interfacing: An Introduction,” Cambridge University Press, USA, 2013.
[16]. Z. J. Koles, “The quantitative extraction and topographic mapping ofthe abnormal components in the clinical EEG,” Electroenc.
Clin.Neurophys.,vol. 79, pp. 440–447, 1991.
[17]. Y. Wang, S. Gao, X.Gao, “Common Spatial Pattern for Channel Selection in Motor Imagery Based Brain computer Interface,”
Proceedings of the 2005 IEEE Engineering in Medicine and Biology 27th
Annual Conference, Shanghai, Chaina, pp. 5392-
5395,2005.
[18]. Kantardzic M., “ Data Mining: Concepts, Models, Methods, andAlgorithms ”, IEEE Press & John Wiley, November 2002.
[19]. M.R.Hasan, M.I.Ibrahimy,S.M.A.Motakabber, and S.Shahid, “Classification of Multichannel EEG Signal by Linear Discriminant
Analysis,” Progress in Systems Engineering: Proceedings of the Twenty-Third International Conference on Systems Engineering,
Advances in Intelligent Systems and Computing 1089, pp. 279-282, 2015.
[20]. Gareis, Ivan E.; Acevedo, Ruben C.; Atum, Yanina V.; Gentiletti, Gerardo G.; Banuelos, Veronica Medina; Rufiner, Hugo L.,
“Determination of an optimal training strategy for a BCIclassification task with LDA,” Neural Engineering (NER), pp. 286 – 289,
2011.
[21]. B. Blankertz, "BCI Competition III", Fraunhofer FIRST.IDA,http://ida.first.fraunhofer.de/projects/bci/competition_iii, 2005.
[22]. G. Dornhege, B. Blankertz, G. Curio, and K.-R. Muller, "Boosting bitrates in noninvasive EEG single-trial classifications by
featurecombination and multiclass paradigms," IEEE Trans. Biomed. Eng.,vol. 51, no. 6, pp. 993-1002, 2004.
[23]. C. Neuper, G. R. M¨uller-Putz, R. Scherer, and G. Pfurtscheller, “Motorimagery and EEG-based control of spelling devices and
neuroprostheses,”Progress in Brain Research, vol. 159, pp. 393–409, 2006.
[24]. L. Zhang, C. Shang, H. Higashi, J. Cao, and T. Tanaka, “Common Spatial Pattern Using Multivariate EMD for EEG
Classification,”Asia-Pacific Signal and Information Processing Association Annual Summit and Conference 2011, October 18-20,
Xi‟an China, 2011.

ElectroencephalographySignalClassification based on Sub-Band Common Spatial Pattern (SBCSP)

More Related Content

What's hot (20)

Similar to ElectroencephalographySignalClassification based on Sub-Band Common Spatial Pattern (SBCSP) (20)

More from IOSRJVSP (20)

Recently uploaded (20)

ElectroencephalographySignalClassification based on Sub-Band Common Spatial Pattern (SBCSP)