B VP, Chinara S. Automatic classification methods for detecting drowsiness using wavelet packet transform extracted time-domain features from single-channel EEG signal.
J Neurosci Methods 2020;
347:108927. [PMID:
32941920 DOI:
10.1016/j.jneumeth.2020.108927]
[Citation(s) in RCA: 15] [Impact Index Per Article: 3.8] [Reference Citation Analysis] [Abstract] [Key Words] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/08/2020] [Revised: 08/01/2020] [Accepted: 08/31/2020] [Indexed: 01/04/2023]
Abstract
BACKGROUND
Detecting human drowsiness during some critical works like vehicle driving, crane operating, mining blasting, etc. is one of the safeguards to prevent accidents. Among several drowsiness detection (DD) methods, a combination of neuroscience and computer science knowledge has a better ability to differentiate awake and sleep states. Most of the current models are implemented using multi-sensors electroencephalogram (EEG) signals, multi-domain features, predefined features selection algorithms. Therefore, there is great interest in the method of detecting drowsiness on embedded platforms with improved accuracy using generalized best features.
NEW-METHOD
Single-channel EEG based drowsiness detection (DD) model is proposed in this by utilizing wavelet packet transform (WPT) to extract the time-domain features from considered channel EEG. The dimension of the feature vector is reduced by the proposed novel feature selection method.
RESULTS
The proposed model on freely available real-time sleep analysis EEG and Simulated Virtual Driving Driver (SVDD) EEG achieves 94.45% and 85.3% accuracy, respectively.
COMPARISON-WITH-EXISTING-METHOD
The results show that the proposed DD method produces better accuracy compared to the state-of-the-art using the physiological dataset with the proposed time-domain sub-band-based features and feature selection method. This task of detecting drowsiness by analyzing the 5-seconds EEG signal with four features is an improvement to my previous work on detecting drowsiness using a 30-seconds EEG signal with 66 features.
CONCLUSIONS
Time-domain features obtained from EEG time-domain sub-bands collected using WPT achieving excellent accuracy rate by selecting unique optimization features for all subjects by the proposed feature selection algorithm.
Collapse