A Finger Movement Recognition Research Based On ECoG Signal Time-freauencv Analysis And Neural Network

The brain is one of the most important organs of the humans.It is responsible for controlling and guiding the important physiological activities of the body.Therefore,understanding its working characteristics and principles is an important research direction.At present,people mainly understand the related activities of the brain through the physiological electrical signals of the brain.By analyzing the electrical signals of the brain,it is helpful to solve the mystery of the working principle of the brain,and then help to research and even develop practical results such as the brain-computer interface system.However,at the same time,EEG signals are also a non-stationary nonlinear random signal that is difficult to study,which is a difficult point for related research.This paper studies the relationship between electrocorticography(ECoG)signal and each finger activity in the public database of brain-computer interface competition.The main content and specific experimental steps of this article are as follows:1.Preliminary introduction background2.Briefly introduce the basic knowledge and processing technology related to ECoG signals.3.Initially use AR model spectrum estimation to estimate the power spectrum of ECoG signals in each channel before and after finger activity,and observe the energy change of EEG signals brought by finger activity.4.Using the method of continuous wavelet time-frequency analysis,calculate the time-varying spectrum of ECoG signals and normalize it,then extract the principle spectrum component(PSC)by principal component analysis and perform low-pass filtering to smooth the PSC waveform,which is used as the feature and recognition result.Then,according to the number of the correlation coefficient,the electrode channel associated with each finger activity is determined.5.The one-dimensional convolutional neural network was used to identify some finger activities,and the resulting identification classification label constitutes a label vector,the feasibility of convolutional neural network for ECoG signal identification was preliminarily verified.At present,the research on ECoG signal recognition is still at a relatively preliminary stage.In this paper,the continuous wavelet time-frequency analysis and convolutional neural network are used to identify the finger-related activities of the human body,and the feature PSC has a great correlation with finger activity.The feature PSC obtained an average correlation coefficient of 0.38 with finger activity and 0.54 in each local finger activity period in the test-set,at same time the identification vector obtained by convolutional neural network has an average correlation coefficient of 0.35 with finger activity also.The result of this article indicate that the PSC can be a effective feature between the ECoG signals and finger activity,and the method proposed provide a reference direction for future related ECoG recognition and brain-computer interface research.