The New EEG Signal Processing Method Based On Low-Rank And Sparse Decomposition

Analysis of EEG data is the basis of Brain Computer Interface,and it plays an crucial role in the field of neuroscience,medicine,psychology and so on.EEG signal processing has become an important topic in BCI.Now the common EEG signal processing methods such as independent component analysis are approximate the original signal through a series of variables from the source of signal.In this paper,we propose a hypothesis from part of signal that EEG signal is superposed by a spontaneous background signal and a mental tasks signal.The inherent background signal is relatively steady signal,and the mental tasks signal is generated by corresponding cortical neurons in certain missions.Under real-world conditions,the collected EEG is sums of above both and the noise.Then,this paper studies the EEG signal processing from the following aspects.1)A model of EEG signal processing based on low rank and sparse decomposition is proposed.Firstly,we introduced the low rank and sparse decomposition and some algorithms in detail,including Iterative Thresholding,Accelerated Proximal Gradient Approach,Dual Approach and Augmented Lagrange Multipliers.Based on low rank and sparse decomposition,we build the EEG signal processing model.It divided EEG into inherent background signal with low rank and mental tasks signal with sparse,and then we explore a new method of EEG signal processing.2)A method of EEG signal processing based on low rank part is proposed.In order to verify the low rank part is background signal which is the relatively steady EEG signal in any case,we identified the subjects based on low rank and sparse decomposition method.It removes the EEG task signal and analysis the background signal of subject.Firstly,we decompose the EEG data after filtering based on GoDec algorithm.The low rank part is used to extract feature based on phase and amplitude information.Then,we combine the low rank part and original EEG signal for another feature extraction.Finally,sparse representation classifier is used for classification.3)A method of EEG signal processing based on sparse part is proposed.In order to verify the sparse part is mental tasks signal which is generated by corresponding cortical neurons in certain missions,we use sparse part for motor imagery experiment based on low rank and sparse decomposition method.It removes the EEG background signal and analysis the tasks signal.Firstly,we decompose the EEG data after filtering based on GreGoDec algorithm and MBRMF algorithm.The sparse part is used to extract feature based on common spatial filtering algorithm.Then,linear discriminant classifier is used for classification.In this paper,we proposed a new method of EEG signal processing based on low rank and sparse decomposition.For the different application we choose the different part for research,which also provides a new idea for brain cognitive science.The experimental results show that the part after decomposition is better than original signal in both two applications.Especially in the theta band which is not good for the original,the identity recognition results of low rank part increased by about 10% and are quite stable.