Research On Key Technologies Of Brain-computer Interface System Based On Walking Imagination

Brain-computer interfaces(BCI)based on walking imagery(WI)belongs to the lower limb motor imagery(MI),which could be used to activate the changes of sensorimotor rhythms for lower limb rehabilitation.Therefore,the research on WI-BCI has important significance.Therefore,the position of the sensorimotor cortex corresponding to the lower limb is deeper than the upper limb,and the area of the sensorimotor cortex representing the lower limbs is smaller than the upper limbs.Therefore,the electroencephalogram(EEG)signals of WI may be more difficult to detect than the upper limb MI and the brain activity patterns of WI are often recognized erroneously.As a result,the classification performance of the WI-BCI is poor,and it cannot be widely applied outside the laboratory,and it has not received much attention.In order to solve the above problems effectively,two research points are paid attention to: how to obtain a effective EEG signal and how to achieve effective classification of WI-BCI.On the one hand,effective EEG signals can enhance the MI-BCI classification effect.WI is a kind of brain-conscious activity controlled autonomously.The stimulus source belongs to the spontaneous source,which depends on the user's own ability.A suitable cue training paradigm could help users improve this ability,and prompt users to effectively modulate EEG signals to obtain reliable EEG signals.In order to achieve a suitable cue training paradigm,I set up virtual environment(VE)and achieved a training paradigm based on VE cue.This training paradigm is used to evaluate the classification performance of WI-BCI.The experimental results show that the training paradigm based on VE cue achieves better performance than the traditional text-based training model and improves the classification accuracy of WI-BCI.This proves that the VE cue can effectively help users modulate brain activity,obtain reliable EEG signals,and improve the quality of EEG signals,and effectively classify WI-BCI.On the other hand,WI-BCI could get the better performance of classification with the state-of-the-art pattern recognition algorithm.Firstly,I use the common spatial pattern(CSP)algorithm to obtain the spatial feature of EEG signals.Secondly,the time-frequency feature of EEG signals are obtained by wavelet packet transform(WLT)and power spectral density(PSD)algorithms.At the same time,the spatial features and time-frequency features of EEG signals are combined to form a multi-view feature(MVF).The experimental results show that MVF improves the performance of WI-BCI classification and can fully express WI EEG information.In addition,deep polynomial networks(DPN)is used to encode single features,and the encoded features are fused to construct multi-view DPN(MDPN).The experimental results show that MDPN improves the classification performance of WI-BCI.In order to better utilize the advantages of DPN,multi-view multi-level DPN(MMDPN)is constructed in this paper.The experimental results show that its classification accuracy is the highest,compared with other deep learning methods,.MMDPN could enhance the complementarity and correlation of multiple features,and further improve the classification performance of WI-BCI.In summary,my research obtains more reliable EEG signals based on the VE-cue training paradigm,and improves the classification performance of WI-BCI based with MMDPN.