Research On Classification Method For Overlapped Spikes

The nervous system of the human supports the function and communication between thevarious organs of the body. The spike produced by the Neurons is the media to transfer anddissemination information between the nervous systems. In the present, then detect the spikeand classify them according the different types of the spike waveform, this process calledspike sporting. Because the limitation of the collection technology, the spike detected will bethe overlapped spikes. The spike waveforms will be different because of the different spike orthe different time they overlapped, which will increase the difficulty of the spike sorting. Sostudying the overlapped spike sorting is the key to increase the classification accuracy, it willbe very significance for the study of neurons coding.This thesis studied the generating principle and the waveform characteristics of theoverlapped spike, analyzed the difficulties of the overlapped spike sorting and research thefeature optimization and spike sorting method of the overlapped waveforms.In order to solve the problem of the information loss caused by the superposition of thespikes, this thesis studied the phase space reconstruction algorithm and applied it to spikesorting, and a new spike sorting method based on the phase space reconstruction is proposed.After phase space reconstruction, the waveform information becomes enrichment, which willhelp to analyze the overlapped spikes from multiple perspectives. This thesis did a lot ofexperiment using the Wave_clus simulation data. The experiments show that this method isvery profit for the overlapped spike sorting, the classification accuracy improved.In order to solve the problem that the phase space reconstruction increased the datadimension and the classification accuracy decreased when the noise interference is serious.This thesis improved the phase space reconstruction. The combined feature optimizationmethod can not only decreased the data dimension but also highlight the waveform trends,which will be helpful to improve the spike classification accuracy. Comparing with themethod only using phase space reconstruction or window-slope representation, theexperiments at different noise level and different superimposed degree shows that thecombined feature optimization method can not only decrease the data dimension but also cankeep high classification accuracy when the noise is serious.