本論文基於共空間模式(Common Spatial Patterns, CSP)與黎曼流形(Riemannian Manifold)切線空間映射(Tangent Space Mapping, TSM)，用於腦電訊號(Electroencephalography, EEG)多分類任務上。藉由濾波器組(Filter Bank)與共變異數矩陣(Covariance Matrix)計算出各個通道間的頻譜功率，再用共空間模式與黎曼流形的切線空間映射分別提取特徵，最後透過基於支持向量機(Support Vector Machine, SVM)的新型分類器分類。資料集使用BCI competition IV 2a想像運動(Motor Imagery, MI)四分類準確率達到78.55%，BCI competition III 3a想像運動四分類準確率達到83.33%，自行錄製之想像運動四分類準確率達到57.44%，自行錄製之實際運動四分類準確率達到81.25%。;This paper is based on common spatial patterns (CSP) and Riemannian Manifold tangent space mapping (TSM) for Electroencephalography (EEG) of multiclass classification tasks. The spectral power between each channel is calculated by the filter bank and the covariance matrix, and then the features are extracted by the CSP and TSM respectively, and finally the new classifier based on Support Vector Machine (SVM) is used to classify. The datasets used BCI competition IV 2a motor imagery (MI) four-classes accuracy rate achieved 78.55%, BCI competition III 3a MI four-classes accuracy rate achieved 83.33%, self-recorded MI four-classes accuracy rate achieved 57.44%, self-recorded motor movement four-classes accuracy rate achieved 81.25%.
