基于卷积神经网络的脑电信号分类

针对现有卷积神经网络脑电信号(electroencephalogram,EEG)分类模型分类精度低、方法复杂且耗时的问题,对卷积神经网络的卷积层进行了改进,提出了多尺度卷积核卷积神经网络(convolutional neural networks,CNN)脑电分类模型,并在输入数据前加了系数矩阵,该系数矩阵可以随网络的训练逐步更新,代替了手工提特征再送入网络的步骤,有助于提高分类精度。最终本文的脑电分类模型在高原脑电信号的分类实验中,二分类准确率比改进前提高8%,三分类、四分类准确率分别达到92.87%、81.15%,分类准确率较高,对脑电信号的分类具有较高的参考价值。

Classification of EEG Signals Based on Convolutional Neural Networks

Aiming at the low classification accuracy, complicated method and time-consuming problem of existing convolutional neural network EEG classification model, the convolutional layer of convolutional neural network is improved, and a multi-convolution kernel convolutional neural network EEG classification model is proposed. And add a W coefficient matrix before the input data, the coefficient matrix can be gradually updated with the training of the network, instead of manually extracting the features and then sending them to the network, which helps to improve the classification accuracy.Finally, the EEG classification model of this paper in the classification experiment of high altitude EEG signals, the accuracy of the two classification experiments is 8% higher than that before the improvement.The classification accuracy of the three classifications and the four classifications reached 92.87% and 81.15%, respectively, and the classification accuracy rate was high, which has a high reference value for the classification of EEG signals.