Abstract

Info

Author(s):: LIU Jie; WANG Raofen; DENG Yuan; School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

Keywords:: Keywords: electrocardiogram signal bidirectional gated recurrent network self-attention mechanism fatigue classification

Abstract:: 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%.