基于STFT的卷积神经网络在MI-EEG分类中的应用  

Application of convolutional neural network based on STFT in MI-EEG classification

作　　者：彭禹 宋耀莲[1] Peng Yu;Song Yaolian(Kunming University of Science and Technology,Kunming 650050,China)

机构地区：[1]昆明理工大学,昆明650050

出　　处：《电子测量技术》2021年第6期36-41,共6页Electronic Measurement Technology

摘　　要：在脑机接口技术中,针对运动想象脑电信号(MI-EEG)由于其自身的高度非平稳性导致传统的手工提取特征困难和分类准确率低的问题,在特征提取和分类这方面进行了研究和探索,设计了一种基于短时傅里叶变换(STFT)的浅层卷积神经网络(SCNN)。通过对网络模型的超参数进行调优实验,该网络模型在2008年BCI竞赛的公开数据集2b上进行测试得到了较好的分类结果,分类准确率达到80.23%,远高于没有进行STFT处理的CNN方法61.04%的准确率。在相同的测试指标下优于传统的机器学习分类方法(CSP+SVM)73.52%的分类准确率,同时也比同类型的深度学习方法(CNN-SAE,77.60%)更具有优势。In the brain-computer interface technology,MI-EEG is difficult to extract features by hand and the classification accuracy is low due to its high degree of non-stationarity.This paper conducts corresponding research and exploration to solve this problem,and designs a shallow convolutional neural network(SCNN)based on short time Fourier transform(STFT).Through tuning experiments on the hyperparameters of the network model,the model was tested on the public data set 2 b of the BCI competition in 2008 and got good classification results.The classification accuracy rate reached 80.23%,which was much higher than the accuracy of the CNN model without STFT processing of 61.04%.Under the same test indicators,it is superior to the traditional machine learning classification method(CSP+SVM)with a classification accuracy of 73.52%,and it also has more advantages than the same type of deep learning method(CNN-SAE,77.60%).

关 键 词：运动想象 卷积神经网络 短时傅里叶变换 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]