Research On Key Technologies Of Emotion Recognition Driven By Multi-physiological Signals

As a multidisciplinary research field involving computer science,psychology and cognitive science,affective computing has great scientific research potential and broad application prospects.Its potential application fields include human-computer interaction,autonomous driving,medical treatment and teaching.Therefore,this field has received extensive attention from researchers.Emotion recognition is an extremely important research direction in the field of emotion computing.Emotion recognition can be achieved based on voice,video,physiological and other signals.Compared with voice,video and other signals,emotion recognition based on physiological signals has the advantages of truly reflecting human emotions in daily life and may overcome possible artifacts masked by human society.Therefore,it has become a hot research direction this year.Although the relevant research has achieved fruitful results,there are still certain limitations:as far as the data is concerned,more focus on a certain physiological signal,the data cannot comprehensively and systematically record and reflect the emotional state of the subject;in terms of characteristics,Focus more on the traditional time-frequency domain features,which cannot characterize the dynamic change of emotional state over time.Based on systematically reviewing domestic and foreign literature on physiological signal-driven emotion recognition,this paper uses the open source data set AMIGOS as a research vehicle to build a multi-physiological signal-driven emotion recognition model.The main research contents of this dissertation are as follows:First,summarizing and analyzing the collection and preprocessing of multiple physiological signals.With investigation and comparison,AMIGOS was selected as the data set for this study.Based on AMIGOS,it summarizes and analyzes how to choose emotion-inducing materials and experimental equipment,how to collect multi-modal physiological signals and the subject's self-assessment report,and how to reduce the noise of the collected physiological signals in the multi-physiological signal collection experiment.Secondly,the time-frequency analysis and non-linear analysis technologies are used comprehensively,and the deep learning technology is combined to construct a feature system.On the one hand,it extracts the characteristics of multiple physiological signals(ECG signal,EEG signal and EEG signal)from three angles of time domain,frequency domain and nonlinear domain;on the other hand,it uses a sliding window to divide the original signal segment analysis;using the deep learning algorithm LSTM to extract the timing characteristics of each sliding window;combining the two aspects to jointly build a feature system.Immediately,the random forest algorithm and the ten cross-validation technique were used for feature selection.Finally,based on the machine learning algorithm,an emotion recognition model is constructed.This study builds recognition models based on four machine learning algorithms:support vector machine,k-nearest neighbor,logistic regression and XGBOOST.After comparing the performance of the four algorithms,the most optimal SVM is finally selected to construct the arousal classification model,and the most optimal XGBOOST constructs the valence classification model.The experimental results show that with the fusion of the features extracted by LSTM and the algorithm,the method proposed in this paper significantly improves the classification performance of emotion recognition;compared with the official experimental results of AMIGOS,the accuracy rate is improved by 23.3%in terms of valence;Just arousal,its accuracy increased by 15.2%.