2021 4th International Conference on Data Science and Information Technology | 2021
Feature Extraction and Classification Recognition of Electroencephalogram Signals Based on Convolutional Neural Network
Abstract
In order to solve the problem of electroencephalogram(EEG) feature extraction and classification recognition, we utilized the EEG activity data captured by the brain-computer interface system, and adopted the deep learning method to construct the training model for visually induced and human spontaneous EEG signals, so as to achieve the classification and prediction results with high accuracy. Firstly, the brain potential data obtained from the 12 target character flicker experiments of 5 subjects from S1-S5 were preprocessed to improve the Signal Noise Ratio (SNR). These included the combined denoising of high and low pass filters successively, the EEG denoising method based on independent component analysis (ICA), the removal of ocular artifacts using EEGLAB Adjust, and the time domain truncation and average superposition to increase the sample size. Secondly, we compared three algorithms: Multi-layer Perceptron (MLP), BP Neural Network, Convolution Neural Network (CNN), built the P300 EEG feature recognition based on CNN and classifier, conducted training and validation for many times. The result shows that compared with other methods, CNN can obtain ideal identification accuracy, and has stable and overall level of higher information transmission rate, so as to verify the rationality of the method of CNN. The P300 EEG feature recognition and classifier based on CNN constructed in this paper can be used for the feature extraction, recognition and classification of evoked EEG signals, which has strong theoretical significance and application value in the process of evoked EEG signal analysis and discrimination.