Research On Eeg Feature Extraction Method Of Steady-State Visual Evoked Potential Based On Multi-Stable Stochastic Resonance

With the increasing maturity of brain-computer interface（BCI）system,the interactive operation of brain-driven external devices has gradually stepped out of science fiction imagination and realized practical application.In BCI systems,steady-state visual evoked potentials（SSVEP）caused by light stimulation are increasingly used as signal sources to determine the user’s true intention.However,there are artifacts and spontaneous electroencephalogram（EEG）signals in SSVEP signals,which can interfere with the effectiveness and correctness of recognition results.Based on this,this paper analyzes the useful information contained in SSVEP signals from the perspective of improving the detection precision,and studies frequency recognition algorithms that can be used to improve the detection accuracy.This research has achieved some results.The specific content is summarized as follows:In order to solve the problem of multiple types of noise in SSVEP signal,which makes it difficult to detect signal characteristic frequency,this paper advances a multistable stochastic resonance（MSR）model.To begin with,giving a minute description of the characteristics of SSVEP signals.Next,deriving the mathematical model of multistable system,including Langevin equation and Fokker-Planck equation.And numerically solve the proposed model using the fourth order Runge-Kutta equation.For large parameter signals,their signal characteristics contradict the adiabatic approximation theory.This paper presents a variable scale solution for them.Finally,the classical EEG feature extraction algorithm,canonical correlation analysis（CCA）,is introduced and applied to the comparative study of the results.To enhance the filtering effect of a stochastic resonance（SR）system,consider introducing a damping factor into the system.With the aid of the second order underdamped feedback principle,the MSR system effectively achieves secondary filtering of the signal.Therefore,the second-order underdamped tristable stochastic resonance（SUTSR）method is proposed.This paper discusses the impact of changes in various parameters of a second-order system（such as damping factor,system parameters）on the output signal-to-noise ratio.By comparing and analyzing the SUTSR simulation results induced by Gaussian white noise and colored noise,it is determined that Gaussian white noise has a stronger excitation effect on the system.Finally,this article conducts an experimental verification of SUTSR and compares it with the detection results of MSR and CCA.The experimental results make known,SUTSR can enhance the extraction effect of SSVEP signal in noise background.Considering the harmonic information contained in SSVEP signals,SUTSR is fully combined with filter banks,and an innovative filter bank second-order underdamped tristable stochastic resonance（FBSUTSR）method is proposed to effectively use harmonic components to enhance signal frequency detection.Using a common data set of 40stimulus targets in 35 subjects,the performance of SSVEP signal characteristic frequency detection using FBSUTSR,SUTSR,MSR,and CCA was compared and analyzed.When the time window is 5s,the FBSUTSR method achieves excellent performance,with an average accuracy of 97.35±3.97%and an average information transmission rate（ITR）of60.58±4.76 bit·min^-1.The results show that the FBSUTSR method has significant performance advantages in realizing SSVEP signal characteristic frequency detection.The selection of parameters of SR system depends on people’s experience,and the parameters determine the output effect of the system.Considering the applicability and superior performance of the system,an adaptive SUTSR method based on particle swarm optimization（PSO）is studied using the output signal-to-noise（SNR）ratio as the fitness function.The optimal SUTSR output is achieved by synchronously optimizing system parameters and damping ratios.Through visual comparison of the output SNR,the output SNR of the system after adaptive optimization is completely higher than that before optimization.The experimental results show that the adaptive SUTSR optimized by PSO can effectively extract the characteristic frequencies of SSVEP signals.