一种基于脑电信号的疲劳驾驶检测方法

A Method for Detecting Tired Driving Based on Electroencephalogram

脑电信号一直被誉为疲劳检测的“金标准”,基于脑电信号的功率谱特征和微分熵特征,构建了基于Conformer和门控Transformer网络(gated Transformer network,GTN)的脑电信号分类器。引入深度可分离卷积,实现了基于脑电信号的疲劳检测方法。该方法使用SEED-VIG数据集进行验证,同时引入了其他主流时序分类模型作为对比。采用所提方法对疲劳状态进行分类时,准确率最高可达97.5%。通过混淆矩阵分析,证明了该方法识别各状态时都有很高的准确率。实验结果表明,微分熵特征在各分类器上的训练效果更好,相比其他模型,所提出的基于Conformer和GTN的分类器在4种特征处理数据集上的平均准确率达到96.2%,具有明显优势。

Electroencephalogram has always been known as the gold standard for fatigue detection.Based on the power spectrum and differential entropy features of EEG,an EEG classifier based on Conformer and Gated Transformer Network(GTN)is constructed.The depth separable convolution is introduced to realize the fatigue detection method based on EEG signal.The method uses SEED-VIG datasets for validation,and introduces other mainstream temporal classification models for comparison.The accuracy of this method is up to 97.5%when classifying fatigue state.Through confusion matrix analysis,it is proved that the method has high accuracy in identifying each state.The experimental results show that the differential entropy feature has a better training effect on each classifier.Compared with other models,the proposed classifier based on Conformer and GTN network has a significant advantage in the average accuracy of the four feature processing data sets,reaching 96.2%.