Research On BCI Data Classification Based On Tensor Decomposition

In the research of brain-computer interface technology,EEG signals are widely used due to its high safety,low price and convenient use.However,the EEG signal has the characteristics of non-stationary,high-dimensional,and low signal-to-noise ratio.How to extract and classify the features of the EEG signal is the focus of BCI research.As a highdimensional data analysis tool,tensor decomposition can combine information from multiple modalities to obtain classification features with discriminative information.Therefore,in this paper,tensor decomposition is used to analyze and study EEG signals.The main research contents and innovations of this article are as follows:Aiming at the problem of high computational complexity in tensor decomposition in high-dimensional space,this paper proposes a tensor R-Tucker decomposition method based on random singular value decomposition with the help of random singular value decomposition for large-scale matrix decomposition.And it is used for feature extraction and classification of BCICIV2 b,a motor imagery-based left-handed classification dataset.The experimental results show that compared with the tensor Tucker decomposition,the feature extraction speed of the tensor R-Tucker decomposition has increased by 22%,and the average classification accuracy has reached 80.93%,which is 10.12% higher than the existing method.It can be seen that tensor R-Tucker decomposition can not only fuse the information of multiple modalities to make the extracted features more discriminative,but also improve the speed of tensor Tucker decomposition,and effectively solve the problem of high computational complexity of tensor Tucker decomposition.In addition,this paper designs and develops an offline BCI data visualization analysis platform using the Python-based Django framework.The platform can analyze EEG data in various formats,and has the advantages of simple and convenient operation,which can conveniently and quickly realize EEG analysis operations such as data loading,data visualization,filtering,segmentation,re-referencing,and time-frequency analysis of offline BCI data.To sum up,this paper proposes a tensor R-Tucker decomposition method,which will be applied to the classification research of BCI data.This method improves the feature extraction speed and classification accuracy of BCI data.In addition,this paper desi gns and develops an offline BCI data visualization analysis system.The platform has the advantages of friendly user interface,simple and convenient operation.