Abstract

Info

Author(s):: WANG Anqi1; YU Chao1; CHEN Yinwei1; XI Qun2; 1.School of Information Engineering, Gansu University of Chinese Medicine, Lanzhou 730000, China 2.Information Center, the Second Hospital of Lanzhou University, Lanzhou 730001, China

Keywords:: Keywords: electroencephalogram emotion recognition deep learning long short-term memory Transformer

Abstract:: Abstract: An electroencephalogram (EEG) emotion analysis model (LTNet) that combines long short-term memory (LSTM) and Transformer modules is proposed for addressing the shortcomings of traditional emotion recognition methods in dealing with long-term dependencies. After data preprocessing, the extracted features are input into LTNet. LSTM module and Transformer module model the input sequence independently, and from which deep local features and global features are extracted and then fused using a weighted fusion strategy. Finally, Softmax function is used to classify emotions into 4 categories. Experimental results show that LTNet has an average recognition accuracy of 96.56% in the 5-fold cross-validation on the DEAP dataset, which is 2.74%-21.31% higher than traditional machine learning algorithms and other deep learning methods.