Study On Emotional Interaction Based On Micro-expression Semantic Cognition

Natural human-computer interaction as an intelligent information technology to promote human life has important research significance.Affective cognitive ability is a key technical index to measure the friendliness of interaction,which has been widely concerned in related research fields.micro-expression,which reflects the real emotional state of human beings,is an important data source of analyzing affective computing models.In this paper,we captured the real emotion of the user with the semantic feature analysis of micro-expression and studied the inner emotional state transfer mechanism of the robot.By learning from the process of "human interaction",we constructed the emotional interaction model by combining individual behavior cognition with psychological cognition,which can enhance the friendliness of human-computer interaction.The main contents and innovations of this paper are as follows:(1)Focusing on visual area of interest,analysis of semantic cognition of micro-expression is carried out,and a method of semantic feature extraction of context based on visual attention is proposed.In the visual area of interest,CNN algorithm is used to extract fine-grained features of micro-expression,and attention model is introduced into spatiotemporal context cognitive module to obtain semantic features of micro-expression.At the same time,sparse coding algorithm is used to realize feature sparseness.Experiments are carried out in CASME2 and SMIC micro-expression databases.The results show that the semantic feature extraction method of micro-expression proposed in this paper can better capture the features strongly related to emotion recognition,and provide important support for emotion mapping of micro-expression.(2)Based on the introduction of emotion information entropy and personalized feature nodes,an incremental emotion mapping model SFT_BLS based on feature transfer is proposed.In the sparse subspace.by introducing the emotional information entropy to analyze the difference of emotion features between domains,a feature transfer algorithm is constructed to improve the accuracy of emotion recognition with small sample size.In the broad learning system,the feature transfer algorithm is used as the input layer,and the personality mapping feature nodes and the personality enhancement nodes are introduced to enhance the personalized cognitive ability of the emotion model.The incremental updating of the model is realized by constructing multi-layer SVD algorithm.Experimental results show that the model proposed in this paper has better recognition rate and can effectively solve the problem of low recognition rate with small sample size and high computational cost with incremental data.(3)The Gross emotion regulation strategy is integrated into the HMM model to construct the emotional state transition model with cognitive ability.According to the Big Five Personality Scale and the first law of emotional intensity,we analyzed the influence of personality characteristics and emotional stimulation intensity on emotional cognition in this paper.By combining gross cognitive reappraisal and expression inhibition with two random processes of HMM model,the probability transfer model between any emotion states is realized,which makes the robot have the ability of human like emotion generation and expression.The experimental results show that the cognitive affective computing model can enhance the robot's affective control ability and promote the friendliness of human-computer interaction.We embed the above emotional interaction theory and model into the robot interaction platform of autistic children,and evaluate the feasibility of the proposed model and its effect on autistic children's auxiliary treatment.In the expression data set of autistic children,the performance of SFT_BLS model in emotion classification of autistic children was observed,and the emotion mapping and tracking of patients' expression in multi-dimensional space were realized by improved GMM model.By tracking the two treatment cycles from 10 patients with mild to moderate autism,it shows that autistic therapy robot with emotional cognitive ability can significantly improve the social response ability of patients.