Влияние выбора стратегии формирования обучающего множества и способа фильтрации на эффективность ИМК, основанного на спектрометрии в ближнем инфракрасном диапазоне

Журнал высшей нервной деятельности им. И.П. Павлова. 2022. Т. 72. № 5. С. 728–738.

Исаев М. Р., Bobrov P.

The paper proposes methods for brain–computer interface, based on the hemodynamic activity registration using near–infrared spectroscopy (NIRS) and adapted for using in the movement disorders rehabilitation. Methods include a filtration adapted to the instructions frequency, step by step classification of the rest state and active tasks, as well as training of the interface classifier on previous sessions data. The effect of the proposed methods and using a smaller set of channels on the accuracy of movement imaging recognition is evaluated on the data of three series of experiments previously done in the laboratory with healthy volunteers. It is shown that the removal of low–frequency noise components from NIRS signal by proposed filtering significantly increases the classification accuracy, as well as session to session learning of the same subject, the number of channels can be reduced without loss of recognition accuracy.

Keywords: фильтрация, filtration, motor imagery, brain-computer interface, интерфейс мозг-компьютер, воображение движений, near-infrared spectroscopy, спектрометрия ближнего инфракрасного диапазона