The Applied Research Of Empirical Mode Decomposition In Local Field Potential Decoding Of Movement

Decoding the movement information of animals' neural signal and predicting animals' movement intention and other parameters are of great significance to revealing neural information coding mechanism of the movement process and neural prosthesis.The neural signals recorded by microelectrode array are usually divided into high frequency spike signals and low frequency Local Field Potentials(LFP).When decoding the neural information,compared with spike,LFP is easy to collect,stable and non-sensitive to time,it can decode the neural information long-term effectively.However the LFP signal is characterized by slow potential changes,non-stationary,large amount of data and are susceptible to noise pollution so that the SNR is quite low,and the traditional signal analysis methods lack adaptive nature to analysis LFP signal.Thus finding a adaptive method to extract feature information of LFP is critical to decode movement information effectively.In the paper pigeons were used as the experimental object,they were trained to complete specific tasks.Then we use the empirical mode decomposition with adaptive characteristics and research the neural information decoding of the movement intention based on LFP of pigeons.We focus on the application of empirical mode decomposition both in the signal preprocessing and feature extraction,which extracts the characteristics information closely related to the training content of the animal behavior from the complex neural signals and applies it to the neural decoding.And then decoding method based on wavelet decomposition to extract the characteristics information of frequency band were compared.The main contents are as follows:1.Experimental design and signal acquisiton.Firstly we designed animal training device and stimulation mode,then we trained pigeons long-term fixed to make them complete the trainings accurately and skillfully.Finally we acquired the neural signals from the movement-related regions of pigeons during the training sessions.2.Preprocessing of LFP signal.EMD and its unique advantages in non-stationary signal analysis and processing were briefly introduced,andthen analysis noise characteristics of LFP signal collected in the experiment.EMD and independent component analysis method(ICA)were combined innovatively for LFP signal preprocessing gathered from each channel,and more to provide pure LFP signal for the subsequent feature extraction and decoding.3.Feature extraction of LFP signal.The feature extraction of the LFP signal began after the preprocessing,which contained the feature extraction based on wavelet decomposition and feature extraction based on EMD.The former decomposed into a series of detail coefficients as a feature set,the latter decomposed into a series of Intrinsic Mode Functions as a feature set.Then time-frequency image and AR power spectrum were combined respectively to the above two methods,and the optimal feature band was chosen with the Fisher discriminant analysis.4.Effect analysis of feature extraction and selection in the movement decoding.Two common decoding algorithms were used to verify the effect of feature extraction based on wavelet decomposition and EMD.Then the two methods were compared with the decoding accuracy and theory to choose an optimal scheme of neural information decoding of movement intention.