Global-to-local feature aggregation deep network method for electroencephalogram emotion recognition

The invention provides a global-to-local feature aggregation deep network method for electroencephalogram emotion recognition. The method comprises the following steps: firstly, segmenting an electroencephalogram signal into a plurality of electroencephalogram signal segments, then filtering the electroencephalogram signal segments through a Butterworth filter, and extracting differential entropy features of the electroencephalogram signal segments; secondly, an undirected graph structure is constructed according to the position relation between the electrodes, and node features in an undirected graph are represented by differential entropy features of the electroencephalogram signals; and finally, node features and connection edge information of the undirected graph are input into a global-to-local feature aggregation deep network model for electroencephalogram emotion recognition constructed by the invention, and the deep network model comprises three parts: a global learner, a local learner and a classifie