Method for improving generalization ability of emotion recognition system

Abstract

The invention discloses a method for improving the generalization ability of an emotion recognition system, and the method comprises the following steps: 1, collecting an electroencephalogram signal: collecting the electroencephalogram signal, and making the electroencephalogram signal into a 310-dimensional vector sample; step 2, establishing an emotion recognition model: firstly mapping 310-dimensional electroencephalogram features into 256 dimensions, then mapping the 310-dimensional electroencephalogram features into 128 dimensions, and respectively inputting four branches: an emotion classification branch, an age regression branch, a gender classification branch and a number classification branch; the gender, age and serial number information of the trainee are input, emotional response is induced through external stimulation, electroencephalogram signals are collected and input into the model to obtain the gender, age and serial number information, and the obtained information and the input information of the trainee are subjected to pairing training; and 4, inputting a to-be-tested sample into the model, and outputting corresponding emotion, age, gender and serial number by the model. The method has the advantages that the cross-testee adaptive capacity of the model can be improved on the premise that a pre-experiment is not carried out, and the generalization capacity of the model can be effectively improved.

Description

technical field [0001] The invention relates to the technical field of EEG emotion recognition, in particular to a method for improving the generalization ability of an emotion recognition system. Background technique [0002] Emotion recognition refers to the recognition of human emotional states, and has broad application prospects in human-computer interaction, health care, security prevention and control, and other fields. Compared with traditional emotion recognition methods based on facial expressions, body movements, speech, text, etc., emotion recognition methods based on non-invasive electroencephalogram (EEG) have higher objectivity, accuracy and stability. However, the EEG signals of different subjects vary greatly, and it is difficult to obtain an EEG emotion recognition model that can be used across subjects. In practice, only labeled data can be collected for each subject (this process is called pre-experiment). , use the Supervised Learning strategy to obtain...

AI Technical Summary

Problems solved by technology

The first type is to use the past data of the same subject to build a model and apply the model to the new data of the subject. The starting point of this method is that the difference of EEG signals of the same subject at different times is low The difference between the EEG of different subjects; the disadvantage of this method is that it cannot handle the case where the subjects have no past data

The second category is to use methods such as Style Transfer to improve the similarity between EEG samples of different subjects, so as to achieve the purpose of sharing classifiers and reducing pre-experiment time; this method can transfer models across subjects , but the disadvantage is that a pre-experiment is still required

The third category is to start from the domain adaptation (Domain Adaptation) method to improve the similarity between the newly collected data and the past data, so as to use the supervision information of the past data when the new data has no labels; this method can improve EEG Cross-subject availability of emotion recognition models, but samples collected in pre-experiments are still required

Images