Emotion Recognition Based On Multiple Physiological Signals

With the rapid development of artificial intelligence,emotion recognition,as an important part of artificial intelligence,has received more and more attention from domestic and foreign experts and scholars.Since EEG signals are the expression of the brain's internal neurons in the cortical tissue,the emotions of EEG signals are also different.It cannot be controlled by any conscious or intentional person.Therefore,it is reliable and feasible to recognize human emotions through EEG and other physiological signals.Sex.Based on the good performance of deep learning in the field of artificial intelligence,relevant neural network algorithms have become the focus of people's research after several decades.People have solved the traditional neural network algorithms through continuous improvement based on traditional neural network algorithms.Problems that are difficult to solve have driven artificial intelligence to a higher level.Based on the existing neural network algorithms,this paper proposes specific improvement measures,and collects emotion EEG signals and heart rate signal data through design experiments to conduct emotion recognition research.The experimental results show that the individual models predict their own emotional recognition accuracy rate is 81.65% on average.,Other models predict their own average accuracy of 87.67%.The main work of this paper is as follows:(1)using chat as a source of emotional stimuli,the brain electricity and heart rate data of the subjects under different emotions(interested,neutral,and uninteresting)are collected,and the raw data is preprocessed using wavelet transform and the like,and the effective signals are Noise is decomposed at different scales.(2)based on the extraction of EEG and heart rate signals in the time domain and frequency domain,based on the traditional features of dimensionality reduction algorithms,this paper proposes a feature dimension reduction method combining single factor analysis of variance and principal component analysis.Theoretical and experimental results show that this method can effectively realize the dimensionality reduction of high-dimensional attributes under the condition of ensuring certain attribute contributions of original effective data.(3)in the analysis of the deficiencies of the LSMT algorithm,an improved LSTM classification algorithm based on decision-making is proposed.And select Google's deep learning open source framework Tensorflow as the emotion model training and emotion recognition framework of the experiment,and conduct emotional learning and recognition analysis on the data after feature reduction.Experiments show that the improved LSTM algorithm can significantly improve the accuracy of experimental results.