Local Temporal Joint Recurrence Common Spatial Patterns for MI-based BCI

As a purely data-driven spatial domain filtering method, common spatial pattern (CSP) is widely used in motor imagery (MI)-based brain-computer interface (BCI) domains. However, the estimation of the covariance matrices in CSP is sensitive to noise and lacks local temporal information. To improve the performance of CSP, we propose a local temporal joint recurrence common spatial pattern (LTJRCSP), which considers temporally local information. Theoretically, LTJRCSP introduces the local temporal joint recurrence rate and therefore may contain more discriminative information and is more robust compared to CSP. After the feature is extracted, linear discriminant analysis (LDA) is then applied for classification. The experimental results, regardless of whether EEG dataset IIa of BCI Competition IV introduces outliers or not, justify the effectiveness of the proposed method.