Research On Motion Information Decoding Based On Forward Potential Signal

Analyzing the information processing mechanism of brain and revealing the relation between brain neural activity and external stimuli,modeling the encode model of neural ensemble to decode the movement intentions has always been one of the forefront of neuroscience research problems.However traditional multi-neuron clusters decoding method using a linear method, although it is simple to use, but the non-linear characteristics of the nervous system is discrepancies, the movement intent on continuous variables, such as high trajectory progress decoding, linear approach would seem stretched. Therefore, this paper take the spike signal as decoding objects studied spike signal detection, feature extraction and classification steps, this article focus on how to build a nonlinear state-space model decoding movement intent and use particle filter algorithm performs decoding tasks.Firstly,given the research satus of decoding the movement intentions from neural singal the article introduce the basic concepts of Bayesian estimation methods, then introduce Kalman filter algorithm and particle filter algorithm from the Bayesian method, at last the article demonstrate the advantages and disadvantages of various nonlinear state estimation methods.In order to obtain the input signal decoding model, we studied the potential pretreatment center spike signal detection and classification of these two aspects. In the detection part, we use morphological filter which is based on the sub-optimization ideas spike denoising original signal, the signal denoised further spike detection extraction. In the classification part, we study the spike signals characteristic which is extract by the principal component analysis method, then use K-means clustering method to classify extracted spike singles, the spike experimental results show that the accuracy of detection is95%and90%,which is more than for the decoding of the nervous system ready.Thirdly, we study the linear optimal estimation and state-space model which is used the static model, the hippocampus neuron spike sequences carry the movement intent decoding rats. We established a model of the linear decoding, then we use two methods decoding the measured signal. The results showed that:he linear decode model has become the bottleneck of the decoding effect, Thus it has become an urgent need to establish consistent with the nervous system to be decoded nonlinear decoding model, and it is used appropriate methods to solve the nonlinear problems to perform decoding tasks. In order to establish the decoding model, we studied the distribution of neuronal receptive field position, given the spike payment reflecting the relationship between activity and exercise position tuning function. Furthermore we have established a state-space form of recursive equations based on particle filtering algorithm using the Monte Carlo method for decoding. We give the use of particle filtering algorithm decoding neural information flow. Finally the particle filter algorithm achieve higher precision linear Kalman filter and optimal estimation doubled decoding result, the correlation coefficient with the real signal is decoded signal reaches0.95. Furthermore, the particle filter can also be achieved accurately decoded signal acquisition under less than ideal conditions, the experimental results show that the particle filter can only use12%the number of neurons of optimal linear estimation and Kalman filter, we can achieve more accurate data decoding to0.85.