基于心电信号的自注意力双向门控循环网络疲劳检测模型

Self-attention BiGRU fatigue detection model based on ECG signal

人类操作员的生理疲劳状态对其作业效率与安全性存在很大的影响,本研究提出了一种基于自注意力(SA)机制的双向门控循环(BiGRU)网络疲劳检测模型,研究基于心电信号的疲劳检测方法。首先采集了模拟不同负荷水平的过程控制任务环境下操作人员的心电数据,以一维心电数据作为输入,经过去噪预处理后,使用改进的BiGRU神经网络进行特征提取,BiGRU在保留GRU优点的同时可以更加充分学习心电信号前后时序的特征联系,并通过SA机制筛选显著相关特征信息,最后将所获得的特征信息经过softmax分类器,得到疲劳分类结果。与传统的GRU模型和BiLSTM模型进行了比较,经过改进后的SA-BiGRU模型的疲劳分类性能整体提高2%~5%,总体准确率达83%。

The physiological fatigue state of human operators has great effects on their working efficiency and safety.A BiGRU fatigue detection model based on self-attention(SA)mechanism is proposed to study the fatigue detection method based on electrocardiogram(ECG)signal.The ECG data of the operators in the process control task environment simulating different load levels are collected,and the one-dimensional ECG data are taken as input.After denoising and preprocessing,the improved bidirectional GRU neural network(BiGRU)which can more fully learn the feature connection of the timing before and after ECG signals while retaining GRU advantages is used for feature extraction,and SA mechanism is used for screening out significant related feature information.Finally,the obtained feature information is passed through the softmax classifier to obtain the fatigue classification results.The fatigue classification performance of the improved SA-BiGRU model is improved by 2%to 5%as compared with the traditional GRU model and BiLSTM model,and the overall accuracy of the improved SA-BiGRU model reaches 83%.