A Source Domain Trial Selection Method for Motor Imagery Based on Resting-State Data Euclidean Space Alignment Transfer Learning
To solve the problem of non-stationarity of EEG signal and complex and time-consuming calibration process,a source domain trial selection method based on Euclidean space rest data alignment is proposed.Firstly,the subjects'EEG trials are aligned with their rest-state data in Euclidean space to reduce the interference of factors unrelated to motor imagery task.Then,according to Euclidean distance measurement criteria,the source domain samples that are far away from the target domain samples are eliminated to further re-duce the difference between the source domain and the target domain samples,so as to improve the effect of transfer learning.The aver-age accuracy rates of 80.71%and 74.46%are obtained on two open motor imagery datasets of BCI Competition IV dataset1 and data-set2a,respectively.The experimental results show that the proposed method can effectively ameliorate the non-stationarity of EEG signal and improve the classification accuracy of motor imagery signals.
国家科技期刊平台
NSTL
万方数据
