并行时间卷积的疲劳检测

Fatigue Detection Based on Parallel Causal Convolution

生物信号是时间序列信息,具有非线性和时变性的特点,疲劳状态能通过脑电信号(EEG)和眼电信号(EOG)等生物信号分析得到。文章提出并行时间卷积神经网络(P-TCN)模型,用于EEG和EOG融合的精神疲劳检测分析,对两种生理信号特征进行BatchNorma处理后输入TCN网络学习,再通过特征融合进行二次学习。使用公开的警觉度SEED-VIG数据集进行实验,结果表明,结合EEG和EOG确实能提高疲劳检测的性能,提出的P-TCN模型与现有的时序处理算法LSTM模型进行对比,不但能学习到生物信号前后的时间信息,同时具有卷积网络并行计算的特点。

Since biological signals are time-series information,they have nonlinear and time-varying characteristics.The fatigue state can be obtained by analyzing biological signals such as electrooculogram(EOG)and electroencephalogram(EEG).Therefore,this paper proposes a Parallel Time Convolutional Neural Network(P-TCN)model for mental fatigue detection and analysis of the fusion of EEG and EOG.First,perform BatchNorma processing on the two physiological signal features,then input the TCN network to learn in parallel,and finally perform secondary learning through feature fusion.This article uses the public alertness SEED-VIG data set for research.The experimental results show that combining EEG and EOG can indeed improve the performance of fatigue detection.The proposed P-TCN model is compared with the existing time sequence processing algorithm LSTM model,which can not only learn the time information before and after the biological signal,but also has the characteristics of parallel computing by convolutional network.