基于内容自适应小波特征与判别式层次原型对齐的跨被试脑电解码网络OA

Cross-subject decoding of motor imagery EEG signals is affected by factors such as non-stationarity and inconsistent feature distributions,leading to insufficient model generalization.To address these issues,this algorithm proposed a cross-subject EEG decoding network(CAWF-DHPA network)based on content-adaptive wavelet features and discriminative hierarchi-cal prototype alignment.The network architecture firstly constructed a content-adaptive wavelet feature extraction module(CAWFE module),which enhanced the stability of spatiotemporal representations by integrating multiscale temporal transient features with relevant frequency-domain information.Subsequently,it employed a hierarchical prototype alignment module(DHPAN module)to achieve collaborative correction of margins and conditional distributions,thereby obtaining feature repre-sentations that possessed both domain invariance and class discriminability.Experiments on the BCI competition Ⅳ 2a and OpenBMI datasets demonstrate that this method achieves average accuracy rates of 79.48％and 77.42％,respectively.The results indicate that the proposed method effectively mitigates inter-subject signal variations,thereby enhancing the stability and generalization potential of MI-EEG decoding in practical deployment.