Research On Emotion Recognition Based On The Multiple Physiological Signals

Emotions are a basic function of the human brain which ensures that humans can survive and adapt.Emotions influence human learning,memory and decision-making.Emotions are also the most direct factor influencing mental health.Considering emotions is thus crucial to the discovery,diagnosis and treatment of mental diseases.With the rapid development of artificial intelligence,human-computer interaction has become a major field of research.Developing systems that can effectively communicate with humans requires giving them the ability to assess user emotions.However,assessing emotions is difficult since people can express their emotional states in many different ways in terms of language,voice intonation,body movements and facial expressions,and people can try to hide their emotions by consciously altering their behavior.To avoid this problem and effectively detect emotions,a promising approach is to analyze physiological signals produced by the autonomic nervous system of humans because these signals cannot be altered to hide emotions.In this work,we study emotion recognition using the ECG,PPG,EMG,SC and RSP signals,and in particular the issue of feature extraction,selection and fusion.To perform feature extraction,we use the ensemble model decomposition(EMD)and Hilbert-Huang Transform(HHT)algorithm.This algorithm can provide instantaneous amplitude and frequency information by using a scaling function and characteristics of the signals to decompose and transform the signals.To optimize the feature set,this work relies on linear fusion to merge the features,and use the information gain ratio to select features.Linear fusion allows combining a variety of features to obtain a comprehensive feature set.In terms of feature selection,this study utilizes method based on the information gain ratio to filter useless features.To verify the effectiveness of the proposed method for recognizing emotions,it was applied on a standard database provided by the Germany Augsburg University.Two kinds of features sets were compared.The first one consists of single type features,while the second one relies on combined features.It was found that the latter is more effective.Moreover,it was observed that when using a single feature set for emotion recognition,the feature set obtained using the HHT algorithm outperforms the other feature sets.For future work,we have recently created a database containing data collected from multiple subjects.Applying the proposed approach on this data yields a more applicable model.