The Study Of EEG Emotion Recognition Based On Capsule Network

At present,affective computing plays a crucial role in artificial intelligence.Electroencephalograph(EEG)has a great advantage in reflecting the emotional state of humans,and has been widely used for emotion recognition.In recent years,more and more deep learning models,especially convolutional neural network(CNN),are applied to EEG emotion recognition.However,many existing CNN-based EEG emotion recognition methods have a complicated manual feature design and feature preextraction stage for the raw EEG signals before network computing,which is actually not in good accord with the data-driven principle of deep learning.More importantly,the internal relationship between different channels of EEG signals is very important for the correct recognition of emotional states.However,the intrinsic relationship among different channels may be neglected by the CNN.To address the above challenges,we study the methods of EEG emotion recognition based on the Capsule Network(Caps Net).The main work is divided two parts:1.We propose a DL framework i.e.,Caps Net for multi-channel EEG emotion recognition.The proposed Caps Net is an end-to-end framework,which can extract features from the raw EEG signals and determine the emotional states simultaneously.More importantly,it can effectively characterize the intrinsic relationship among various EEG channels.We conduct experiments on two public datasets,i.e.,DEAP and DREAMER.The proposed method achieves state-of-the-art performance.2.We propose MLF-Caps Net on the basis of Caps Net for multi-channel EEG emotion recognition.In comparison to the original Caps Net,the MLF-Caps Net can combine multi-level features extracted from different convolution layers to form primary capsules,which can enhance the capacity of representation of capsule network.In addition,we add a bottleneck layer to reduce the amount of parameters and accelerate the speed of calculation.We also conduct experiments on DEAP and DREAMER.The results show that compared with the accuracy of the method based on Caps Net,the accuracy of the proposed method based on MLF-Caps Net is improved.