A Novel Real-Time EEG Based Eye State Recognition System
| Autor: | Zijia Zhou, Jianqi Liu, Weikuo Dong, Pan Li |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
medicine.diagnostic_test
Computer science business.industry 0206 medical engineering Feature extraction Wavelet transform Feature selection Pattern recognition 02 engineering and technology Electroencephalography 020601 biomedical engineering Random forest Wavelet 0202 electrical engineering electronic engineering information engineering medicine 020201 artificial intelligence & image processing Artificial intelligence business |
| Zdroj: | Communications and Networking ISBN: 9783030061609 ChinaCom |
| Popis: | With the development of brain-computer interface (BCI) technology, fast and accurate analysis of Electroencephalography (EEG) signals becomes possible and has attracted a lot of attention. One of the emerging applications is eye state recognition based on EEG signals. A few schemes like the K* algorithm have been proposed which can achieve high accuracy. Unfortunately, they are generally complex and hence too slow to be used in a real-time BCI framework such as an instance-based learner. In this paper, we develop a novel effective and efficient EEG based eye state recognition system. The proposed system consists of four parts: EEG signal preprocessing, feature extraction, feature selection and classification. First, we use the ‘sym8’ wavelet to decompose the original EEG signal and select the 5th floor decomposition, which is subsequently de-noised by the heuristic SURE threshold method. Then, we propose a novel feature extraction method by utilizing the information accumulation algorithm based on wavelet transform. By using the CfsSubsetEval evaluator based on the BestFirst search method for feature selection, we identify the optimal features, i.e., optimal scalp electrode positions with high correlations to eye states. Finally, we adopt Random Forest as the classifier. Experiment results show that the accuracy of the overall EEG eye state recognition system can reach 99.8% and the minimum number of training samples can be kept small. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|