Modified BBO-based Graph Convolutional Recurrent Neural Network for Emotion Recognition

With the development of emotional intelligence technology, emotion recognition based on electroencephalogram (EEG) has become a hotspot in affective computing. However, since EEG signals are multi-channel signals, there may be high redundancy between different channels. Removing redundant channels can avoid unnecessary noise interference, reduce the complexity of data processing, and maintain the key features of the signal. Graph convolutional recurrent neural networks (GCRNN) can simultaneously obtain the signal's spatiotemporal information, but setting appropriate model parameters is also a thorny problem. Therefore, we propose a graph convolutional recurrent neural network model based on a modified biogeographic optimization algorithm for EEG emotion recognition tasks, which can simultaneously realize model hyperparameter search and EEG signal channel selection and improve the model's ability to perform emotion recognition tasks. The modified biogeography-based optimization (MBBO) algorithm combines biogeography-based optimization (BBO) and the differential evolution algorithm (DE), thereby improving the algorithm's exploration ability and convergence speed. It more effectively selects suitable channels and optimizes network parameters. The proposed model can automatically discover the relationship between input channel features and model parameters. Experimental results show that the proposed method performs well on multiple EEG signal datasets.