基于卷积网络的运动想象脑电自制数据集分类算法研究

Research on Classification Algorithm of Self-Made Data Set of Motor Imagination EEG Based on Convolutional Network

运动想象脑电信号分类已成为脑计算机接口研究领域的一个热点。本文通过实验室设备进行脑电采集并制作自己的数据集,同时将卷积神经网络与传统方法进行结合,提出了一种基于短时傅里叶变换和连续小波变换对原始数据进行特征提取使用卷积神经网络进行分类的算法。利用特征提取算法提取时频特征制成时频图并使用卷积网络快速学习特征进行分类。试验结果表明,该算法在运动想象脑电公共数据集中有着96%的准确率,在自制数据集上准确率达到92%左右,证明了该算法在运动想象脑电分类上的可行性。

Motor imaging EEG signal classification has become a hot spot in the field of brain computer inter-face research. This paper uses laboratory equipment to collect EEG and make its own data set. At the same time, it combines convolutional neural networks with traditional methods to propose a feature extraction method based on short-time Fourier transform and continuous wavelet trans-form. Convolutional neural network classification algorithm uses feature extraction algorithm to extract time-frequency features to make time-frequency map and uses convolutional network to quickly learn features for classification. The test results show that the algorithm has an accuracy rate of 96% in the motor imagery EEG public data set, and the accuracy rate is about 92% on the self-made data set, which proves the feasibility of the algorithm in motor imagery EEG classification.