چكيده به لاتين
Abstract:
Brain computer interfaces (BCI) were introduced to help people who lost their neural channels between brain and muscles. BCI decodes the ongoing neural activities and translates them into output commands that accomplish the desire of the user. There are a variety of methods to record and decode brain activities of which the EEG signals are more favourable in terms of ease of establishment, non-invasiveness and lower cost. Among the EEG-based BCI, the systems which use sensorimotor rhythms are chosen because of the minimum dependence on the performance of other body parts.
Two main requirements for increasing usability and user acceptance in motor imagery based BCI systems are: 1) short training time, and 2) auto-adaptation in the online calibration. In the first part to overcome the problems of training time, based on subject independent methods, we introduce a new design that uses a combination of previous and new user data to form the model. In the second part that is related to the online phase, we propose to eliminate some noisy parts of the data from the training set in orther to further improve the performance of the classification. The results showed that by implementing the proposed auto-adaptive methods, an average increase in the classification accuracy from %75.8 to %86.92 achieved. In conclusion, the methods presented in this study include using the previous users’ data for preliminary calibration, conditioning training set and auto-adaptive online calibration, improved the BCI system in terms of calassification accuracy.
Keywords:Brain Computer Interface based on motor imagery, sensorimotor rhythms, Auto adaptive calibration, subject independent training, weighting data, short training time.
