Research Research On Modeling Approaches Of Emotion Recognition Based On Weighted Fusion Strategy

As robotics advanced,the human-computer interaction tasks have increased and complicated.The cognition of emotion capability of robot can be realized based on the emotion recognition model.Emotion recognition models performance includes operations,recognition rates and so on.During the modeling process,feature engineering,model analysis and model fusion involve fusion of multichannel information.Therefore,in order to improve the recognition rate,the research on modeling approach based on weighted fusion strategy has important theoretical significance and engineering value.This paper takes emotional state of human as the research object and improves the recognition rate as research goal.The research of modeling approaches is developed as emphases.Given the diversity of type of emotion signal and emotion feature,and recognition rate and calculations of recognition model as requested,the key technologies such as multi-modalities feature extraction strategy,weight determination method and weighted fusion strategy are studied in depth for modeling.The main work of this paper is as follows.1)Facial expression recognition based on feature level fusion.With the research results of facial structure and psychology,facial behavior changes are strongly linked to not the whole face but some specific areas,such as eyebrows,eyes,mouth and nose.We extract 2-dimensional coordinate of facial key points as geometric feature.Setting deep learning as breakthrough points,we design a structure based on CNNs for feature extraction orderly and independently.Then deep feature can be obtained which more representative of the sample.According to characteristics of facial expression recognition model,feature level fusion strategy is introduced for series feature vectors.This strategy enhances complementarity of multi-modalities information,thus improve the recognition rate of modal.2)Facial expression recognition based on modal level fusion.According to facial structure,we select feature point which can reflect facial morphology steadily.The whole face can be divided into several areas based on motion range of facial muscles.Feature points divided in teams based on facial areas.Then,we introduce feedback principle into weight calculation of each area feature according to magnitude of recognition rate.And we introduce principle of rigid as a basis for testing weight.And we introduce principle of rigid as a basis for testing weight.At last,according to characteristics of facial expression recognition model,nonlinear weighted fusion strategy is founded by weighted kernel function.This strategy enhances the influence of strong correlation feature and reduces the influence of weak correlation feature by nonlinear weighted feature,thus improve the recognition rate of modal.3)Emotion recognition based on decision level fusion of multiple physiological signals.Firstly,various physiological activities of the human body are related to emotion state,and correspond to various types of representation signals.We select EEG,ECG,RSP,GSR and extract emotion feature of each physiological signal respectively.Then,we introduce feedback principle into weight calculation of each set of multimodal emotion feature according to magnitude of recognition rate.It is motivated by the fact that the expression of multimodal emotion signal on various emotions is different.At last,according to characteristics of emotion recognition model,decision level weighted fusion strategy is founded by principles of max-win.This strategy takes the full advantage of each physiological signal effectively,thus improve the recognition rate of modal.4)Emotion recognition based on feature level and decision level fusion of multimodal emotion signals.According to the complexity of body structure,we select sequences of facial expression images as visual signal and EEG,ECG,RSP,GSR as physiological signal to form multimodal emotion signals.Facial expression feature and one kind physiological feature are concatenated together to form 4 sets of multimodal emotion feature based on feature level fusion strategy.Then,we introduce feedback principle into weight calculation of each set of multimodal emotion feature according to magnitude of recognition rate.It is motivated by the fact that the expression of multimodal emotion signal on various emotions is different.At last,according to characteristics of emotion recognition model,decision level weighted fusion strategy is founded by principles of max-win.This strategy takes the full advantage of each multimodal emotion signal effectively,thus improve the recognition rate of modal.