Трофимов Александр Геннадьевич

pp. 114-136

The paper offers a developed method of constructing a feature space for electroencephalogram classification. It is based on the localization of brain’s electrical activity sources. The simplest statistical characteristics of dipole moments for equivalent current dipoles are chosen as features for classification, and the nearest neighbour algorithm is used for classification. The research on real electroencephalograms reveals that the accuracy of the proposed method is comparable to the accuracy of the existing classical approaches in brain-computer interfaces at the same time giving a number of opportunities to further increase it and having clear neurophysiological interpretation.

The problem of multidimensional non-stationary signal classification is considered. The method of signal feature vector construction based on analysis of dynamical patterns is described. The neural network approach is proposed to use for classification in feature space. The results of experimental researches of developed algorithms for applied problem of electroencephalograms recognition are described.

Method for dynamical systems identification using self-organizing neural networks is considered. Using method VQTAM was constructed self-organizing neural model for blood glucose level of 1-st type diabetic patients. Set of experimental researches for constructed model is performed y is estimated its accuracy