Classification And Recognition Algorithm Of Motor Imagery Based On EEG And Application In BCl

As a new means of human-computer interface, brain-computer interface technology based on EEG has become a research hotspot in the fields of rehabilitation engineering and biomedical engineering. Brain-computer interface technology doesn’t rely on normal output channels like peripheral nerves and muscles but builds a direct communication channel between the human brain and external environment. The correct classification of EEG data is a key factor to determine the performance of brain-computer interface, so the study of EEG-based classification and recognition algorithms has great practical significance.This thesis takes brain-computer interface based on motor imagery as the object of research, carries out a systematic study on feature extraction methods and classification algorithms of EEG based on motor imagery, and realizes an online BCI system using sliding window ICA as the core algorithm. The main content of the thesis is as follows:1. Design the BCI experimental paradigm based on motor imagery and collect abundant EEG data of left and right hand motor imagery. Combining with the existing standard EEG data, a real-life EEG signal database for offline BCI analysis is established.2. Pattern recognition methods such as posterior probability support vector machine(PPSVM) and hidden markov model (HMM) are studied and compared under the EEG pattern classification problem. The "idle" state in motor imagery is detected by PPSVM to create good conditions for online motor imagery control system.3. Study and implement EEG feature extraction methods based on energy, common spatial patterns and sliding window ICA algorithm. A new envelope extraction method based on sliding window ICA is emphasisly studied, which achieves a higher recognition rate in dynamic characteristic analysis of mu rhythm and motor imagery classification. 4. Design and implement an online BCI system based on sliding window ICA algorithm, the experimental results demonstrate that the left and right hand motor imagery can be identified on-line by this system and the recognition rate is up to92.7%.