P300 Detection with Adaptive Filtering and EEG Spectrogram Graph

A new method of P300 potential detection is given in the paper. Firstly, by creating the signal differential profile, a single electrode choose method of P300 is given. Secondly, aiming at the shortcoming of traditional smoothing method, adaptive filtering method is applied to the signal preprocessing of P300 potential for the first time in this paper. This method takes the average signal of the original data as the reference signal and can effectively eliminate all kinds of noise components mixed in the P300 potential. Furthermore, different from traditional time-domain feature extraction method of P300 potential, the paper proposes a feature extraction method based on short-time Fourier transform, which fully extracts the frequency-domain features of P300 potential while retaining the time-domain features. Finally, a convolutional neural network classifier is designed and obtains an average classification accuracy of 89.81% in the different average times of reference signals which was about 5% higher than traditional methods.