摘要
针对稳态视觉诱发电位(SSVEP)脑电信号存在个体差异性强、信噪比低等特点而导致其识别困难等问题,提出一种用于SSVEP信号分类识别的深度学习方法。该方法以原始多通道SSVEP信号为输入,利用SSVEP信号的时空特性,首先使用一维时间卷积核对输入信号的时域进行卷积操作;然后使用一维空间卷积核进行空域卷积,对多通道信息进行融合;最后采用降采样、多尺度卷积、全连接等操作完成SSVEP信号的分类识别。实验结果表明:利用该方法在较短时间的视觉刺激下即可实现对被试者SSVEP信号的有效识别;在1 s刺激时长时,该方法的平均离线信息传输率为94.17 b/min,平均识别准确率为93.3%,相比于无监督典型相关分析方法和有监督支持向量机分析方法,识别准确率分别提升了48.73%、41.21%。该方法具有较高的目标识别效率及鲁棒性,有效提高了基于稳态视觉诱发电位信号的脑-机接口的性能。
A deep learning method for target recognition in the steady-state visual evoked potential(SSVEP)-based brain-computer interface(BCI)is proposed to solve the problems that the strong inter-subject variability and low signal-to-noise ratio existing in SSVEP signal make target difficult to be identified.The proposed method utilizes the original multi-channel SSVEP signal as an input and makes full use of the spatial-temporal characteristics of the SSVEP signal.Then a one-dimensional time convolution kernel is used to convolve the input signal in time,and the spatial convolution is performed by using a one-dimensional spatial convolution kernel to fuse the multi-channel information.Finally,the classification of the SSVEP signal is accomplished by operations such as downsampling,multi-scale convolution,full connection,etc.Experimental results show that the proposed method effectively identifies the SSVEP signal in short response time for each subject.When a one-second time window is applied the average offline information transfer rate(ITR)and the average recognition accuracy of the proposed method are 94.17 b/min and 93.3%,respectively.Comparisons with the unsupervised canonical correlation analysis(CCA)method and the supervised combination method of CCA with support vector machine(SVM)show that the recognition accuracy of the proposed method improves by 48.73%and 41.21%,respectively.It is concluded that the proposed method has superior target recognition efficiency and robustness,and could effectively improve the performance of the SSVEP-based BCI.
作者
杜光景
谢俊
张玉彬
曹国智
薛涛
徐光华
DU Guangjing;XIE Jun;ZHANG Yubin;CAO Guozhi;XUE Tao;XU Guanghua(School of Mechanical Engineering,Xi’an Jiaotong University,Xi’an 710049,China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2019年第11期42-48,共7页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(61503298)
关键词
稳态视觉诱发电位
脑-机接口
目标识别
深度学习
steady-state visual evoked potential
brain-computer interface
target recognition
deep learning