Independent component analysis neural network

مقدم أطروحة جامعية

al-Nayami, Auns Qusayy H.

الجامعة

الجامعة التكنولوجية

الكلية

-

القسم الأكاديمي

قسم هندسة الحاسوب

دولة الجامعة

العراق

الدرجة العلمية

دكتوراه

تاريخ الدرجة العلمية

2004

الملخص الإنجليزي

The complex system such as human brain generates electrical activity from thousands of neurons in the brain.

This activity is given as electroencephalogram ; (EEG) waveforms. The EEG recordings consist of electrical potentials in many different locations on the surface of the skull's skin (scalp).

EEG potentials represent the combined effect of potentials from a fairly wide region of the cerebral cortex.

These are very gross types of summation of potentials from an extremely large number of neurons in the vicinity of the sensors (electrodes) that are used to pick up the EEG activity.

These potentials are presumably generated by mixing some underlying components of brain activity.

The mixing of brain fields at the scalp is basically linear mixture. The present research aims to design and implement an unsupervised neurocomputing.

mode! for separating the original components of brain activity waveforms from their linear mixture.

This is called the problem of "Blind Source Separation" (BSS).

it recovers the unobservable original independent sources from several observed (mixed), data masked by linear mixing of the sources, when nothing is known about the sources and the mixture structure. The neurocomputing model was implemented using the recently developed source separation method " Independent Component Analysis " (ICA) technique for solving blind EEG source separation problem.

This ICA is used to decompose the observed data into components that are as statistically independent from each other as possible.

A neural-network model is proposed, and corresponding unsupervised learning algorithms are developed to achieve the separation. Two types of ICA algorithms have been used for linear BSS problemr they are: ♦ " Nonlinear Principal Component Analysis " (NPCA) algorithm. ♦ " Fast Fixed Point " (FFP) algorithm. The performance and effectiveness of the proposed model was tested lag-computer simulation.

This is achieved by applying conventional Waveforms, such as: sine, cosine, square, saw-tooth...etc waveforms and Scibimbination of these waveforms in many examples. Then after, the performance of the proposed method was investigated using real EEG data signals obtained from normal and abnormal states from the (Neurosurgery Hospital) in Baghdad.

Normal states includes some mental task, and abnormal states include epilepsy disorder only. The proposed ICA neurocomputing model was implemented using the Matlab version6.1 package. The results of the present work show that the ICA can be effectively used to separate the EEG signals from their linear observation records.

Separated EEG signals has been playing a key role in the diagnosis of some brain diseases. Simulation results highlighted the good performance of the proposed model in separating the mixed signals.

Moreover, the separation of real EEG data signals, i.e., the independent components of the model proved to be an important mean of presenting a new data to doctors and physicians.

This is approved by the physicians of neurosurgery.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

الموضوعات

• الشبكات العصبية(الحاسبات الإلكترونية)

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-305907