A Fusion Algorithm for EEG Signal Processing Based on Motor Imagery Brain-Computer Interface

Electroencephalogram (EEG) signal processing is a very important module in the brain-computer interface system. As an physiological feature of human body, EEG signals are closely related to functional state cerebral nervous However, collected on scalp generally weak and inevitably subject various noise interferences. In order remove artifacts from brain computer interfaces (BCIs), fusion algorithm for preprocessing proposed. The method includes following steps: firstly, raw separated into set statistics independent components (ICs) by improved FastICA algorithm. Then, each component decomposed series intrinsic mode functions (IMFs) using empirical decomposition (EMD). Many IMFs with high-frequency deleted. rest reconstructed. Furthermore, further eliminated iterative process algorithm, then, reconstructed again inverse ICA. Finally, cleaned was obtained. comparative experiment shows that EMD-ICA not only accurately eliminates artifact but also better retains local characteristics EEG. Continuous wavelet transform used extract energy features μ rhythm id="M2"> β rhythmic represent under different motor imageries. These two normalized as input data convolutional neural network (CNN) designed paper, kinds learned CNN, two-classification problem imagery completed. experimental results show average classification accuracy kappa value proposed higher than those SVM SAE most subjects.