Analyzing EEG For Emotion Recognition Based On Deep Learning

Emotions are an indispensable part of everyone.It affects all aspects of our work and life to a large extent.Therefore,letting machines recognize human emotions correctly is an obstacle that must be overcome in the field of artificial intelligence.The study found that objective physiological signals such as electroencephalogram,electrooculogram,myoelectricity,etc.can more accurately reflect people's inner emotions than subjective information such as facial expressions,behaviors,gestures and so on.Among them,the EEG signals in objective physiological signals are generated by the central nervous system of the brain and are closely related to the generation and recognition of emotions,so analyzing human emotions through EEG signals has become a current research hotspot.Emotion recognition based on EEG signals is a typical machine learning problem.The core of the problem includes signal feature selection,feature extraction and classification model design.When selecting features,in the field of traditional signal processing,especially for non-stationary signals,short-term energy and short-term amplitude are commonly used features.However,both the short-term energy and the short-term amplitude only focus on global information rather than local information within a certain period of time,which does not reflect the position information of the signal well and causes signal distortion.When using multi-channel EEG signals for emotion recognition,traditional machine learning methods not only rely on knowledge of relevant domains such as time,frequency and space domains to design and extract features from each channel,but traditional machine learning classifiers are generally shallow models.The ability to learn data features is weak and the recognition rate is low.In response to the problems in the current research,this paper combines classic digital signal processing methods and deep learning methods to achieve emotion recognition.The innovative results achieved are as follows:First,when selecting features,short-term amplitude spectrum features are used to replace short-term energy features and short-term amplitude features to represent signals.The short-term amplitude spectrum is the representation of the signal in the frequency domain,which can eliminate the abnormal sensitivity to the larger value signal,and pay more attention to the details than the short-term energy and short-term amplitude,which is beneficial to retain all the information of the original signal.Secondly,this article discusses the application of deep learning methods in the field of EEG signal emotion recognition,and proposes a feature extraction algorithm based on deep learning,and on this basis,a combination algorithm combining feature extraction and feature classification: using convolution The neural network performs feature extraction and the deep belief network performs feature classification.Deep learning methods do not require researchers to learn various relevant domain knowledge in advance to obtain the required features,thereby simplifying the processing.In addition,deep learning methods use features learned from massive data to overcome the problem of feature redundancy and better reflect the inherent nature of the data.Finally,this paper experimentally verified the proposed method on the DEAP dataset.The experimental results show that for four emotion recognitions,the feature extraction algorithm proposed in this paper can achieve an accuracy rate of 93.37%,and the proposed combined algorithm of feature extraction and feature classification can achieve a recognition accuracy rate of 97.46%,which is better than the current existing result.