Emotion Recognition Based on Selected EEG Signals by Common Spatial Pattern

As a more reliable method for assessing human emotional states, electroencephalogram (EEG) signals have been widely used for emotion recognition. In this article, an EEG-based emotion recognition method is proposed, which uses a common spatial pattern (CSP) to extract and select features from the electrode channels. To reduce the redundant data and select key features, the nonparametric test is applied to select a subset of electrode channels and frequency bands from the SJTU emotion EEG dataset (SEED). The test evaluates the differential entropy (DE) features of each channel, resulting in six distinct channel subsets distinguished by their test outcomes. Furthermore, we identify a specific frequency band optimized for effective emotion recognition. The proposed approach is deployed on each subset and the CSP features are obtained through spatial filtering for feature selection. A method of batch normalization (BN) is used on the selected CSP features to mitigate the influence of individual differences on emotion recognition. The performance of the normalized CSP features is then assessed by using ten classical classifiers for emotion recognition. The best recognition accuracy is 88.89% for the selected electrodes and 85.92% for the gamma frequency band. The application of BN enhances the recognition accuracy across each channel subset. Notably, according to the nonparametric test, the CSP features obtained by our method exhibit significant distinctions and are further improved by the application of BN. These results underscore the effectiveness of our proposed method for emotion recognition using only 13 electrode channels and one frequency band.