Classification of Normal and Epileptic Seizure EEG Signals Based on Empirical Mode Decomposition - Complex System Modelling and Control Through Intelligent Soft Computations - page 368

2.3 Feature Extraction

Feature extraction is an important step in pattern recognition and plays a vital role in
detection and classi cation of EEG signals by extracting relevant information.

Feature extraction can be understood as

finding a set of parameters which effectively
represent the information content of an observation while reducing the dimension-
ality. These parameters explore the property of two classes which has separate range
of values for different classes. Two different area measures which are related with the
variability of the signal are used here as a feature set. These area measures are
computed for first four IMFs to create feature vector space. Final feature set consists

of eight features for classi cation of normal and epileptic seizure EEG signals. The

computation of these area measures have been described in detail as follows:

2.3.1 Analytic Signal Representation and Area Computation
of Circular Region

The IMFs that have been obtained using EMD method on EEG signals are real
signals. These IMFs can be converted to analytic signals by applying the Hilbert
transform.
Analytic signal of x ð t Þ can be de ned as (Huang et al. 1998 ; Lai and Ye 2003 ):

z ð t Þ ¼ x ð t Þþ jy ð t Þ ð 4 Þ

where, y ð t Þ represents the Hilbert transform of the real signal x ð t Þ ,de ned as

follows:

1
p

y ð t Þ ¼ x ð t Þ

t

Z 1

ð 5 Þ

1
p

x ðsÞ

¼ p v d s

: :

t s

1

with Fourier transform

Y ðxÞ ¼ j sgn ðxÞ X ðxÞ ð 6 Þ

where p : v : indicates the Cauchy principle value, and X ðxÞ is Fourier transform of

signal x .

The signal z

ð t Þ

ð t Þ can also be expressed as:

e j /ð t Þ z ð t Þ ¼ A ð t Þ ð 7 Þ