EEG detection and de-noising based on convolution neural network and Hilbert-Huang transform

Electroencephalogram(EEG) is the signal fulling of randomness and non-stationarity. It's very susceptible by a variety of noise, especially electrooculogram (EOG). In order to reduce experimental errors, it is necessary to perform artifact recognition and de-noising on the acquired original signal. On the basis of the traditional methods, this paper presents a method of artifact detection and remove based on convolution neural network(CNN) and Hilbert-Huang transform(HHT). Firstly, the instantaneous power of the EEG signal was calculated. The CNN model was used to extract features. The softmax classifier was used to classify EEG. Then, empirical modal decomposition is employed to the EEG with artifacts. The noise in the high frequency component is filtered by referring to Hilbert transform spectrum. Finally, the residual signal is separated by FastICA method to remove the EOG. The experimental results show that the accuracy of CNN method is over 80%. The EEG signal is more pure after HHT de-noising. This work lays a good foundation for the follow-up study.