Research And Application On Emotion Recognition Of Employee Based On Multi-source Physiological Signal With Data Fusion

The development of emotion recognition and work state management of employee has become a new era demand of enterprise management.The development of data fusion technology and portable intelligent collection equipment has provided technical support.Emotion recognition has also shown application value in many fields and accumulated practical experience.This paper takes the realization of rapid and accurate emotion recognition as the research problem,and develops the research of emotion recognition model based on multi-source physiological signal data fusion,in order to apply it to employees’ daily emotion recognition and work status feedback in enterprise management.In this paper,the multi-source physiological signal data is used as the model input,and the one-dimensional convolutional neural network-support vector machine(1-D CNN-SVM)emotion recognition model is constructed to extract the emotional features of the multi-source physiological signal data,and realize feature-level data fusion and complete emotion recognition to output four-category of emotion under the valence-arousal two-dimensional model.The main research work and results of this paper are summarized as follows:(1)Trial construction of an emotion recognition model using the laboratory dataset DEAP.Firstly,data preprocessing and dimensional splicing are performed on the multi-channel physiological signal data to realize different data splicing combinations.Based on the feature-level data fusion method,the emotional features of each part are extracted by1-D CNN model fusion respectively.Then,feature splicing is performed and the output layer and Softmax function in the 1-D CNN-Softmax model are replaced by the SVM classifier to realize the classification and recognition of emotions.Through the comparative analysis of experiments,the model parameter settings of 1-D CNN and SVM are determined,the advantages of the 1-D CNN-SVM emotion recognition model structure are confirmed,and the optimal data splicing combination EEGa&EEGb&PHY is found,and the highest recognition accuracy is 94.79%.(2)Using the daily psychophysiological data set DAPPER as the application data,the simulation demonstrates the application effect of the 1-D CNN-SVM emotion recognition model in this study in the daily working and living conditions of employees,and the experiment confirms the advantages of multi-source physiological signal data fusion and the validity of this model,and the recognition accuracy of the simulation application reaches82.05%.Finally,the paper puts forward the thought of supporting employee emotion recognition-feedback-care cycle mechanism and management application.