Research On Emotion Recognition Based On Improved CN

Emotion is an important expression of human beings to external stimuli.In recent decades,research on emotion recognition has become an important research field,and emotion recognition based on EEG signals is one of its hot research directions.EEG is a general reflection of the electrophysiological activity of nerve cells in the cerebral cortex.It can directly reflect an individual’s emotional state and has a positive effect on the recognition of human emotions.In traditional research,EEG features are mainly time-domain,frequency-domain and time-frequency domain features,and are extracted manually;At the same time,related research generally faces the problem of low accuracy of emotion recognition.Due to the continuous improvement of computer computing efficiency and the rapid progress of deep learning technology,the deep learning method is gradually applied to emotion recognition and has achieved good recognition results in emotion recognition.Therefore,in order to improve the accuracy of EEG recognition,this paper focuses on the deep learning network algorithm in the process of emotion recognition.The main work of this paper includes the construction of classification model and feature fusion.Firstly,the collection experiment of EEG signal stimulated by video is designed,and the wavelet threshold denoising of EEG signal is carried out at the same time,and the differential entropy,power spectral density and wavelet entropy are extracted.The DE feature is used as the basis of deep learning through comparative experiments.In this paper,three models are constructed to classify EEG features,and the specific process is as follows:Firstly,build the DE-CNN model.The DE features of EEG signals were extracted and emotion recognition was performed in a convoluted neural network classification model.The experimental results show that the classification accuracy of the CNN model based on DE features is 3% higher than that of the original CNN model,indicating that the CNN model based on DE features is more accurate.Secondly,construct the DE-LSTM model.Extract the DE feature of EEG signal and input it into the long-term and short-term memory model for emotion recognition.The experimental results show that the accuracy of LSTM model based on DE features is higher than that of CNN model based on DE,which is about 6% higher,indicating that LSTM network has better performance in processing temporal EEG features.Thirdly,construct the DE-CNN-LSTM model.In view of the limitations of CNN in EEG signals,a deep learning model of convolutional cyclic neural network CNN-LSTM based on DE characteristics is designed.The network fully combines the ability of feature extraction and time series information expression of two deep learning networks.Because a single feature may lose some EEG information,multi-feature fusion is carried out.The features before and after fusion are input into the model for comparison,and it is found that the recognition accuracy rate after fusion is 90.9%,which verifies the effectiveness of the new model in the emotion recognition task.