Research On Detection And Feature Extraction Algorithm Of Neuron Spike Signal

As for brain information processing mechanism and brain-computer interface(BCI)research,it is of great significance to extract brain wave signals related to physiological information or behavior from the brain.In order to obtain more accurate brain wave signals and obtain effective brain information,the way of recording the extracellular microelectrode array of neurons is rapidly developed,and the recorded action potential(spike potential)can provide more accurate signals and finer control.The spike potential of the implanted BCI are increasingly attracting the attention of researchers.In the process of extracting brain wave signals,the neural signals generated by the recording of extracellular microelectrode arrays of neurons are usually used to obtain the spike potential.However,the generated spike potential is a non-stationary brain wave signal whose amplitude is smaller,the signal-to-noise ratio is lower,and noise interference signals are easily introduced during the acquisition process.In addition,there are neuroelectric signals emitted by a plurality of other neurons around the electrodes,forming overlapping spikes.These will greatly hinder the detection and feature extraction of the spike waveform.In the analysis and processing of the spike signals,the detection and classification of the spike waveform plays an increasingly important role in the research of encoding and decoding of neural information.In order to be able to analyze and process neural signals more accurately,it is crucial to improve the accuracy and efficiency of spike signal detection and classification.Aiming at a large number of noise signals and the problems existing in the process of overlapping spikes,this thesis has carried out research and improvement from three aspects:spike detection,feature extraction and cluster analysis.The specific work contents are as follows:1)Two classical spike detection methods-threshold detection and peak detection-are analyzed and applied to the spike detection.The results were compared by software simulation,and the advantages and disadvantages of the two waveform detection methods were analyzed.In view of its shortcomings,this thesis adopts the waveform detection algorithm based on mathematical morphology,which firstly uses the mathematical morphology method to denoise the original spike waveform,and then uses the classic peak detection algorithm to detect the spike.This improved method effectively reduces the number of spikes for missed detection and false detection,and improves the accuracy of detection.2)In the feature extraction process,the implementation methods of waveform feature analysis and principal component analysis are analyzed,and the results are obtained through software simulation.Although both methods can achieve the selection of the feature of the spike waveform,many important feature information is still ignored in the process of extracting the waveform feature quantity.In view of the above problems,this thesis combines with wavelet principal component analysis based on wavelet transform.This method can extract the main feature quantities in the spike waveform more effectively,and the accuracy and efficiency of feature extraction are higher.3)In the cluster analysis of neuron spikes,the classic k-means clustering algorithm,template matching method,maximum and minimum distance method and feature-based classification method are analyzed.The simulation analysis is carried out by software.pointing out the advantages and disadvantages of these methods.In view of the above deficiencies,this thesis adopts a cohesive hierarchical clustering algorithm and applies it to the classification of spike waveforms,and compares it with the previous experimental results.It can be seen from the comparative analysis that the clustering using agglomerative hierarchy The algorithm can effectively improve the accuracy of the spike classification.