EEG based Brain Signal Acquisition and Analysis I

• Main Author: Lim, Li Wei
• Other Authors: Ser Wee
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/67983
• Institution: Nanyang Technological University

Electroencephalographic (EEG) is a widely recognized device in medical field. It reads and collects electric signals produced by human brains. Useful information can be analyzed from the collated brainwave signals. In this project, the main objectives are to gather useful signal patterns which translate into two emotion states – happy and sad; to explore different features from the collected signals; to discover a reliable signal feature that can significantly represent the two emotion states; and to develop an algorithm to automatically classify the features selected. Features are selected using Fisher’s ratio. The classifier assigned to this project is linear discriminant analysis. The ability to read and to accurately classify signals as happy signals or sad signals reaps indisputable benefits in medical field. This discovery contributes to better understanding of the complex electric signals produced by human brains. This project involves data collection, data processing, data analysis and data classification. In data collection, an experiment is designed to collect three types of signals, namely resting state signal, muscle artifact signal and emotion stimulated signal. The bandwidth of the brain signal selected for analysis are alpha and beta signal, residing in the frequency ranging from 8 Hz to 30 Hz. The features explored in this project are discrete wavelet transform (DWT) for both approximate and detail coefficient up to level 5, approximate entropy and sample entropy. Before extracting the features for analysis, the signals are filtered at 50 Hz for power line interference noise removal and normalized. During classification, repeated k-fold method is adopted to ensure a stable and consistent classification result. From the classification, result shows that DWT at level 1 gives the best accuracy, with an average of 73.63% accuracy score. However, both entropies give an average of less than 50 % accuracy score. DWT at level 1 provides a more distinct classification between happy and sad signals.