An artifact removing method fusing FastICA and CNN for EEG signal
Mengzhe Yue, Xiaozhong Geng, Linen Wang, Xi Zhang · 2022
Electroencephalogram (EEG) is susceptible to interference from various noises during the acquisition process. In order to remove artifacts and improve classification accuracy, an artifact removal algorithm that combines FastICA and convolutional neural networks is proposes. First, the Convolutional Neural Network (CNN) is used to detect artifacts by calculating the voltage amplitude of the EEG signal, and then the EEG signal containing artifacts is decomposed into independent components by the FastICA algorithm based on negative entropy; Then the threshold method is used to automatically identify the artifact components, and the identified artifact components are removed by FastICA; finally, a pure EEG signal is obtained through reconstruction. Experimental results prove that the fusion method can effectively remove artifacts while retaining useful EEG signals, and ultimately improve the classification accuracy.