基于迁移学习和残差网络的SSVEP信号识别  

SSVEP Signal Recognition Based on Transfer Learning and Residual Networks

作　　者：尹菁 王贤敏[1,2] 王力哲 郭海湘[2,3] YIN Jing;WANG Xian-min;WANG Li-zhe;GUO Hai-xiang(Institute of Geophysics and Geomatics,China University of Geosciences(Wuhan),Wuhan Hubei 430074,China;Key Laboratory of Geological and Evaluation of Ministry of Education,China University of Ceosciences(Wuhan),Wuhan Hubei 430074,China;School of Economics and Management,China University of Geosciences(Wuhan),Wuhan Hubei 430074,China)

机构地区：[1]中国地质大学(武汉)地球物理与空间信息学院,湖北武汉430074 [2]中国地质大学(武汉)地质探测与评估教育部重点实验室,湖北武汉430074 [3]中国地质大学(武汉)经济管理学院,湖北武汉430074

出　　处：《计算机仿真》2024年第9期323-329,共7页Computer Simulation

基　　金：国家自然科学基金(U21A2013,71874165);地质探测与评估教育部重点实验室主任基金(GLAB2020ZR02,GLAB2022ZR02)。

摘　　要：针对脑电信号中的稳态视觉诱发电位(SSVEP)信号目标识别难以适应个体差异、识别稳定性差、精度低的难题,提出了一种参数共享迁移学习的残差网络SSVEP信号识别方法。首先,利用离散小波变换将多通道SSVEP信号转化为小波系数,并与变换前信号构成特征矩阵作为输入特征集,提升特征提取的丰富性;其次,建立融合空间注意力机制的残差网络,利用清华大学脑—机接口提供的两个SSVEP信号数据集,包括105名被试,进行跨任务的迁移训练,把源域上训练完成的网络逐模块迁移至目标网络以获取合适的迁移模块,迁移后连接2层残差块和模式识别单元得到跨个体差异识别结果。实验结果显示,在1s时间窗口,训练与测试使用被试无交集情况下,测试集的总识别率达到84.2%,提升了脑电信号识别的个体适应性,验证了提出的方法在提高SSVEP信号识别的稳健性和准确性上具有优势。A residual network SSVEP signal recognition method based on parameter sharing transfer learning is proposed to address the challenges of adapting to individual differences,poor recognition stability,and low accuracy in target recognition of steady-state visual evoked potential(SSVEP) signals in EEG signals.Firstly,the multi-channel SSVEP signals are transformed into wavelet coefficients using discrete wavelet transform as the input feature set together with the pre-transformed signals;thus,the extracted features are more abundant.Secondly,a residual network fused with a spatial attention mechanism is established,and two SSVEP signal datasets,including 105 individuals,provided by the Tsinghua University brain-computer interface are used to achieve a cross-task and cross-individual transfer.The network trained on the source domain is migrated to the target network block by block to obtain the appropriate transfer block,and the recognition results are obtained by connecting the 2 residual blocks and pattern recognition units after the transfer.The total recognition rate in the test set reaches 84.2% under a 1s time-window with no intersection between training and test individuals.Thus,the proposed method is characterized by relatively high individual adaptability,accuracy,and robustness in SSVEP signal recognition.

关 键 词：稳态视觉诱发电位 残差网络 迁移学习 注意力机制 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]