Abstract: Emotional calculation is studied by means of electroencephalogram (EEG) signals. Aiming to solve the problem of difficulty in extracting EEG signal features and building large computational models, an innovative method is proposed for obtaining reliable distinctive features from EEG signals. The feature extraction method combines differential entropy with linear discriminant analysis (LDA) and can be applied to feature extraction of emotional EEG signals. Three types of emotional EEG data sets are used to conduct experiments. Experimental results show that the feature extraction method can effectively improve the performance of the EEG classification: the average accuracy is improved by 68% compared to the results of the original data set, which is 7% higher than the result obtained by feature extraction using only differential entropy. The total execution time indicates that the proposed method has a lower time complexity. The research results have important practical significance in the field of three-category EEG emotion recognition and can be effectively applied to the actual engineering field.