Electroencephalograph-based emotion recognition using convolutional neural network without manual feature extraction

Electroencephalograph (EEG) based emotion recognition has been studied for a long time with the rapid development of brain–computer interface and electrode techniques. This paper introduces a new EEG-based emotion recognition model built with the convolutional neural network (CNN) to classify three emotions: positive, neutral, and negative. The proposed method adjusts the convolution kernels of the CNN to adapt to the EEG input signals. Unlike the manual feature extraction used in the traditional methods, the proposed method constructs the CNN with direct EEG data input, which ensures the integrity of information utilization and achieves better accuracy at 86.10% on average. In addition, based on the research on full-channel and full-band data, four different profiles of 4, 6, 9, and 12 channels and five frequency bands are selected to study the critical factors that affect the recognition results.