SVM-based Multi-classification for Detection of Vigilance Levels with Single-Channel EEG Signals

Low-vigilance driving behavior is an important cause of frequent traffic accidents, and automatic detection of vigilance levels of drivers is extremely meaningful. To classify four different vigilance levels (awake, semi-awakening, drowsy and sleep state), we designed a driving fatigue experiment and a multi-class classifier based on support vector machine (SVM). Firstly, the EEG signals of different vigilance levels are decomposed and reconstructed at seven levels using the Daubechies 4 wavelet (db4) transform method. Then, extract the standard deviation (S), amplitude logarithm (L), quartile (Q) and coefficient of variation (CV) from the EEG signals and the corresponding decomposed sub-band signals, and construct the feature vectors. Feature vectors are input into the multi-class SVM classifier to classify different vigilance levels. By comparing the classification results of different features of the vigilance states, it is found that the best results are obtained when using the d5 sub-band signals for classification. The combination of channel PO4 and P05 of occipital region has the best classification accuracy in the SVM classifier when the feature is CV, CV + L or CV + Q, and the classification accuracy can reach 99.61%. Furthermore, even if only one channel PO4 and one feature CV were adopted, we can also get relatively ideal classification accuracy 99.41%. Therefore, the proposed classifier can accurately identify four classes of vigilance levels, reduce the computational complexity, and make the detection system more efficient and practical.