The emotion classification of traditional brain network uses clustering coefficients, average shortest path and other topological attributes as classification features. To solve the problem that EEG emotion recognition based on these attributes is susceptible to network connectivity thresholds and attribute selection, and network topology attributes are difficult in fully characterizing the differences in network structure for different emotional states, this paper proposed an EEG emotion classification based on common spatial patterns of brain networks topology(EEC-CSP-BNT). EEC-CSP-BNT calculated the functional connectivity matrix in the sensor space using mutual information for each sub-band, and employed the CSP to learn the spatial filters and constructed classification features. At last, it employed the pattern classifiers(such as Fisher linear discrimination, support vector machine and K nearest neighbor) to complete the emotion recognition. Experimental results using DEAP and SEED datasets validate the superior performance of EEC-CSP-BNT compared to the network topology attributes features. EEC-CSP-BNT can also extract the useful classification information of brain network topology.

[2021-05-05] Printed Article /doc/2020.07.0181/v2