الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 170 |  |  |
| | | from | 170 | | |
Abstract
Brain machine Interface (BMI) is a direct communication channel between brain and computer. It provides communication methods between the brain signals that are detected from the scalp Electroencephalogram (EEG) and an external device (robot arm, wheelchair…). People who are paralyzed or have other sever movement disorders need alternative methods for communication and control. It allows them to control the environment without the need to control muscle activity. We show that multi-electrode recordings from the primary motor cortex can be used to predict arm and hand muscle activity. The Electroencephalogram (EEG) signals that come from the scalp containing an electrical signals bearing what is going on in the human brain or the things that would like to do it. Current BMIs suffer from many problems including inaccuracies, delays between thought, false positive detections, and constraints on invasive technologies that need cost money and cause many dangerous during surgery to the subjects.
The purpose of this thesis is to examine this area and to build solutions for these problems. It designs and implements an offline brain computer interface, and translates the human arm’s movements into commands to simulator robot arm to execute the desired movement. Different techniques have been tested to reduce the number of recordings where not all electrodes distributed over the whole scalp are useful in BMI systems that are known as channel selection. Other techniques have been tested in preprocessing stage to filter and reduce unwanted data to gain the features extraction from the brain signals. More than technique was used for feature extraction such as wavelet transform (WT) that was the most appropriate choice to use time–frequency domain. Furthermore a comparison between these methods was made to evaluate the best methods for feature extraction techniques. WT gave the best results within data classification using Back propagation Neural Network (BNN) and achieved classification accuracy reached to 91.1 %. Normalization procedure was introduced that augmented the performance and accuracy of neural networks classification, which reached to 92.2%. All the techniques that have been tested compared favorably with work done by others in this field