Research On Mental Workload Classification Algorithm Based On ECG Co-dimension Features

Nowadays,the working scenes of human-computer interaction are more and more abundant.Identifying the workload level of the operator's brain and giving timely warning can effectively reduce the incidence of accidents,ensure life safety and improve work efficiency to a certain extent.Electrocardiogram(ECG)can not be artificially controlled and forged,and the acquisition of ECG is safe,cheap,noninvasive,easy to use,and will not affect the daily operation of personnel engaged in dangerous types of work.The technology of automatic ECG signal analysis is currently an important topic in the classification of mental workload.However,due to individual ECG signal differences and noise impact,it is still faced with many difficulties to identify it accurately and efficiently.Based on this,the main research and analysis work of this thesis are as follows: First,with regard to ECG signal preprocessing,aiming at the sudden abnormalities in ECG signal,this paper proposes a frequency correlation method according to the similar frequency domain composition of each heartbeat of ECG signal.This method accurately and automatically eliminates the interference signal segments without manual intervention by setting the threshold value,which lays the foundation for subsequent feature extraction and mental workload classification,It also provides a new idea for other similar ECG signal processing.Wavelet transform soft threshold denoising is applied to ensure the overall quality of ECG signal.Combined with notch filter and differential threshold method,accurate R-wave position is obtained,which lays a foundation for subsequent feature extraction.Secondly,in terms of ECG feature extraction,this paper extracts the features of ECG signals under low,medium and high mental workload levels,and extracts 14 features in total.It includes the more mature time domain,frequency domain and nonlinear features of Heart Rate Variability(HRV)in the field of mental workload research,as well as the innovative frequency domain feature TP wave power,QRS complex power and nonlinear feature sample entropy of ECG waveform extracted from the research of cardiac pathology and EEG.In view of the different feature extraction cycles of HRV features and ECG waveform features,A co-dimension ECG feature extraction method of sliding window and resampling is proposed,which maximizes the use of the features in mental workload classification.Finally,this paper establishes the three classification mental workload model of ECG co-dimension features based on the random forest algorithm by comparing the classification accuracy and test time of the classification model of k-Nearest Neighbor,Support Vector Machines(SVM),Decision Tree,Random Forest and Ada Boost algorithm,as well as the classification accuracy obtained by combining the 14 dimensional ECG features into 8 different inputs according to the types,The average classification accuracy is 96.406%.