Research On Brain Fatigue Mechanism And Sleep Stages Recognition Based On Complex Network Multi-information Fusion

Brain-computer interface(BCI)technology establishes effective pathways for information exchanging between brain and external environment.BCI has been developed with rapid speed and applied to many fields,such as intelligent wheelchair,intelligent robot and smart home system.BCI-based smart home system has attracted a great deal of attention in recent years.Smart home system can be controlled by the instructions extracted from EEG signals.However,after a long use of operation,the fatigue will occur,which will affect the application effect and user experience.Besides,how to realize and improve autonomously sensory ability of the BCI-based smart system remains to be investigated.The interaction between user and smart home represents a challenging problem of significant importance.Aiming to solve the above challenging problems,this dissertation designs a BCI experiment interface based on P300 paradigm.Subjects capture the target image in the stimulate interface to induce ERP potential.External devices including a computer,a television,a lamp,a curtain,an air condition and a water heater can be controlled by identifying ERP potential characteristics.This dissertation records the EEG of several subjects when they are in normal state and fatigue state.The results show that the fatigue will lead to a decrease of the P300 classification accuracy.In order to probe into the reasons,this dissertation develops a multivariate weighted recurrence network to construct weighted brain networks from EEG in fatigue and normal state,respectively.This dissertation employs the average weighted clustering coefficient and weighted global efficient to charactierze the derived networks,aiming to reveal the fatigue behavior from BCI-based smart home system.An increase trend in clustering coefficient in fatigue is found,which is indicative of tight coupling between the corresponding electrode regions.These results suggest that,to some extent,when in fatigue state,brain try to reduce cognitive load by reinforced company and synchronicity.Human-machine hybrid intelligence is the advanced form of artificial intelligence.In order to realize the effective interaction between user and smart home,this dissertation develops a multilayer limited penetrable visibility graph to detect sleep states.In particular,we construct limited penetrable visibility graphs from EEG signals recorded from different subjects under stages of wake,shallow sleep and slow wave sleep,and then,we compute 7 indexes including clustering coefficient entropy,average betweenness,average clustering coefficient,average degree,global efficient,transitivity and average edge overlap to characterize the topological properties of network.Then we take the generated network measures as feature a vector to input support vector machine(SVM)for monitoring different sleep states,and the classification accuracy reaches to 96.95%,which indicates that our method can effectively identify three kinds of sleep stage and allows improving the autonomously sensory ability of BCI-based smart home system.