Recognition of Epilepsy EEG by Two-View Adversarial-Incentived-Based T-S Fuzzy Classifier

The multi-view fuzzy learning method, due to its ability to comprehensively explore feature information and possess good interpretability, has been widely used in the field of disease-assisted diagnosis. This study intends to apply two-view learning techniques to the recognition of epilepsy electroencephalogram (EEG) signals. Given that existing two-view epilepsy detection methods hardly ever consider the adversarial relationship between features obtained from different feature extraction methods, and seldom utilize this relationship for adversarial learning, this study proposes a hierarchical TSK fuzzy classifier (TA-HTSK) based on two-view adversarial learning for epilepsy EEG signal recognition. The decision information and the approximation capability of the trained view in the previous layer is utilized as the adversarial incentive for the current layer's view, providing positive guidance for the training of the view in the current layer. The intra-view and inter-view consistency in the previous layer, as well as the intra-view consistency in the current layer, are considered to improve the stability and recognition accuracy of TA-HTSK. Additionally, this study proposes a stack-like input set construction method to ensure high interpretability of the model. Finally, the excellent performance of TA-HTSK on two-view epilepsy datasets demonstrates the effectiveness of the proposed fuzzy modeling strategy based on adversarial relationships between two views.