Research On Emotion Recognition Method Based On Deep Learning And Brain-Computer Interface For Service Robot

The recognition and classification of emotions have been deeply studied in the academic field and some cutting-edge achievements have been made.However,no matter in the aspects of two-dimensional image or three-dimensional facial expression,or in the field of speech recognition,the emotion recognition rate that can be achieved at present is a little low.This thesis adopts the deep learning models with outstanding performance in recent years,and adds some improvement and innovation.The experiments are conducted from three directions of vision,speech and electroencephalogram respectively.After that,these models in the thesis are combined with a service robot to make the robot have an emotion recognition function.Firstly,in the field of vision,Capsule Net and deconvolution image reconstruction errors are used to train the Cohn-Kanade Dataset(CK+),which is about facial expression and has been enhanced.This model can be used to classify seven emotions,namely anger,disgust,calm,joy,fear,sadness and surprise.By comparing this model with the other famous deep learning models,it can be concluded that this model in the thesis gets the best performance in accuracy and convergence speed.Secondly,in the field of speech,Bi-LSTM(Bi-directional Long Short-Term Memory)is adopted,and the attention model(AM)is added to pay more attention to the more important and prominent features extracted.Then the dropout mechanism is used to reduce the network complexity and avoid over-fitting.The experiments are conducted on EMODB,which is divided into seven emotions.The experimental results are compared with the results in the paper,and it is found that the stability and the highest accuracy of the recognition rate obtained by the model in this thesis are much better.Finally,an Electroencephalogram(EEG)cap with 8 electrode channels is selected to collect EEG data.After the EEG cap and the detection software of the EEG on computer is connected through the brain-computer interface(BCI),the machine learning model can be used to conduct the experiments about the emotion classification of the data.After completing the above three types of emotion research experiments,the training model is connected with the NAO service robot to realize a simple emotional interaction process between the service robot and the human beings.