A Study On Spike Detection And Classification Method Based On Quaternion

Spike is the carrier of nervous system to receive, analyze and communicate information. Brain processes and integrates the information transmitted in the sequence to restore the feeling and response. Consequently, spike detection and classification is the foundation of researching neural information encoding, which has very important significance of revealing the working mechanism of the brain. Multichannel electrodes can record multiple spike sequence simultaneously, which provide us with a window on realizing how neurons work together. But records of the signal which have large amount and complex form bring new challenges to neurons signal processing. In this paper, according to the characteristics of multichannel spikes, the quaternion signal processing methods are applied to multichannel spike preprocessing, detection, feature extraction and classification.According to the characteristics of spike data, combined quaternion differentia l filtering synchronization de-noising method with wavelet threshold de-noising, we propose a new method of the parallel pretreatment of multichannel spike. Based on the simulation data and international public neurons spike database, the proposed method is verified to restrain various spike noise and baseline drift, which can reduce waveform diversity of similar spike and contribute to the exact classification of spikes.According to quaternion algebra theory, this paper proposes spike sphere threshold detection algorithm based on quaternion root mean square and implements its visualization in three-dimensional space. In dealing with simulated and real spikes,quaternions threshold detection algorithm in operation time and the results of detection are better than traditional threshold method and window threshold method.Spike feature extraction is the precondition of achieving its accurate classification. Because of large amount of data in the actual spike potential classification experiments, the existing multi- channel spike feature extraction method dispose slowly, and lose correlation information between each channel data. In this paper, according to the structure characteristics of multichannel data, the improved quaternion principle component analysis based on the quaternion bidiagonalization is used to extract the characteristics of multichannel spike, which can fuse with multi-channel data information and reduce the computational complexity of QPCA. Aiming at quaternion characteristics, quaternion K-means clustering algorithm based on quaternion distance is proposed. Compared with normal K- means clustering results of serial PCA characteristics, the proposed spike parallel processing is confirmed to be superior to traditional serial processing, and similar to artificial classification. Thus, the algorithm can improve the efficiency in the actual spike classification instead of artificial classification.