Research On High-speed Rail Driver Vigilance Detection Based On Multi-physiological Signal Features

High-speed railway(HSR)promotes social and economic development as well as facilitate people’s travel,and in recent years their safety has become an increasingly widespread concern.As the emergency manager of trains under the CTCS-3 level of the Chinese Train Control System(CTCS),the HSR driver is responsible for ensuring the safe operation of trains.However,the monotonous and high-pressure driving tasks can easily lead to distraction and fatigue,and prevent the driver from taking over and controlling the train in case of emergency,endangering the safety of the train.The mobility and convenience of wireless wearable technology allows for the collection and wireless transmission of human physiological data anytime and anywhere.To address this background,this thesis collects a variety of physiological signals from HSR drivers through wireless wearable devices and builds a classification model based on deep learning to detect the alertness of HSR drivers.The main research contents of this thesis are summarized as follows:Firstly,after reviewing and studying the appropriate electrode acquisition locations,a wireless wearable Open BCI device was used to acquire Electroencephalogram(EEG),Electrocardiogram(ECG)and Electromyography(EMG)signals from the subjects at different states of vigilance.The data were also pre-processed using linear filtering,fast independent component analysis and wavelet threshold denoising in order to eliminate noise generated during data acquisition and to ensure data reliability.The pre-processed data is labelled with dataset labels depending on the state of acquisition,and self-made training and test datasets are created.Secondly,due to the specificity of physiological signals,this thesis uses a onedimensional convolutional neural network to build the SE-Res Net V2 model for channel attention mechanism and the Fa-Alex Net model for specific multi-physiological signal alertness detection.The models were trained and tested on a self-made dataset,and finally achieved 97.47% and 96.14% accuracy on the test set respectively.Finally,the SE-Res Net V2 model and the Fa-Alex Net model evaluated in comparison using classification evaluation metrics,and the effectiveness of the added SE module and different combinations of physiological signals on the performance improvement of the network model was verified in both models by ablation experiments respectively,and the detection model built in this thesis was compared and analyzed with existing methods based on physiological signal alertness detection.