Fatigue Prediction Of Twin Neural Networks Based On Bimodal Bioelectricity

Fatigue driving has serious safety hazards,so the research on driving fatigue prediction has very important practical significance.Traditional fatigue monitoring methods mostly use a single mode for fatigue classification,which has problems such as low classification accuracy and easy to be affected by the driving environment.In view of the existing problems,this paper uses the bimodal physiological signal of EEG signal and upper limb EMG signal to predict fatigue,and classify the level of fatigue.First,a fatigue-inducing experiment was designed.EEG and EMG signals were collected through Emotive epoc+and NeuSen WM,respectively.Four channels were selected for EEG signal collection;EMG collected triceps long head,triceps lateral head triangle The signals of the six-pack muscle,infiaspinatus,subscapularis and latissimus dorsi were compared,and the changes of the two physiological signals of EEG and EMG before and after fatigue were compared.Secondly,the interference noise of EEG and EMG signals is removed,and three features of sample entropy,mean and power spectral density are extracted as EEG fatigue features;three features of intermuscular consistency,average power frequency and integral EMG value are extracted as EMG fatigue characteristics.A 600*30 brain muscle fusion feature dataset was obtained through feature fusion,which contains 300 awake data and 300 fatigue data.Finally,the fused data is screened by improving the RanSac algorithm.The algorithm finally screens out the most representative 161 1*30-dimensional fatigue data points,and takes the average of these 161 data points to define the fatigue standard value.As a standard value input of the Siamese neural network,the data to be tested is used as another input of the Siamese neural network.The features of the two sets of input vectors are extracted through the convolutional neural network,and the output results are limited by the sigmoid function.The results are divided into three fatigue classes.In order to verify the feasibility of the Siamese neural network to predict fatigue,CNN_LSTM and SVM were selected to classify the data.The results show that the classification accuracy of CNN_LSTM and SVM is about 75%,while the classification accuracy of the Siamese neural network has been greatly improved,which is higher than the traditional classification method.Therefore,driving fatigue prediction based on twin neural network has certain research value and practical significance.