叠加锋电位的分类方法研究

发布时间：2018-06-10 20:13

本文选题：锋电位分类 + 叠加锋电位　；参考：《天津理工大学》2014年硕士论文

【摘要】：人体神经系统支撑着体内各个器官之间的功能及联系。神经元发放的锋电位是神经系统间传递和发布信息的媒介。目前常采用多电极胞外记录的方式对锋电位信号进行采集，从采集的信号中将锋电位信号检测、提取出来，并对不同形态特征的锋电位信号进行各自归类，这一系列的过程就称为锋电位分类。受当前采集技术的限制，当多个神经元细胞在相隔很短时间同时放电时，检测到的会是多个锋单位信号的叠加波形。叠加锋电位的波形形态会因为叠加时刻和叠加波形的不同而不同，这增加了锋电位分类的难度。因此，研究叠加锋电位的分类，是提高锋电位分类准确性的关键，对神经元集群编码的研究具有重要意义。本文通过对叠加锋电位的产生原理及波形特点进行研究，重点分析了叠加锋电位分类的难点问题，研究适合于叠加锋电位识别的特征表示方法和分类策略。为解决锋电位波形由于叠加而造成的波形信息丢失问题，本文通过对相空间重构算法进行研究，将其应用到叠加锋电位分类中，提出了基于相空间重构的锋电位分类策略。通过对锋电位进行相空间重构，丰富了其波形信息，可以从多角度对叠加锋电位样本进行分析。本文的方法，在Wave_clus数据集上进行了大量实验，实验结果表明，基于相空间重构的锋电位分类策略对叠加锋电位分类具有良好的效果，准确率有了一定程度的提升。针对基于相空间重构的分类策略中数据维度大及抗噪声能力差的问题，本文对其进行了改进，在相空间中引入了基于窗口斜率表示法和二阶差分表示法相结合的特征表示方法。结合之后的特征优化方法，不仅可以降低锋电位样本的维度，还可以将锋电位波形的变化趋势突出，有利于提高锋电位分类的准确性。通过对不同噪声水平和不同叠加程度的锋电位数据进行实验，结果表明，，与单独用相空间重构算法及窗口斜率表示法相比，结合之后的特征表示方法不仅在一定程度上降低了数据的维度，而且对噪声干扰严重的锋电位数据能够保持较高的分类准确率。
[Abstract]:The human nervous system supports the function and connection of various organs in the body. The spike potential sent out by neurons is a medium for the transmission and dissemination of information between the nervous systems. At present, the multi-electrode extracellular recording is often used to collect the spike potential signal. The spike potential signal is detected and extracted from the collected signal, and the spike potential signals with different morphological characteristics are classified respectively. This series of processes is called spike classification. Limited by the current acquisition techniques, when several neuron cells discharge simultaneously at very short intervals, the superposition waveforms of multiple front unit signals will be detected. The waveform shape of superposed spike is different because of the superposition time and superposition waveform, which makes it more difficult to classify the spike. Therefore, the research on the classification of superposed spike potential is the key to improve the accuracy of spike potential classification, and it is of great significance to study the coding of neuron cluster. In this paper, the generation principle and waveform characteristics of superposed spike potential are studied. This paper mainly analyzes the difficult problems in the classification of superposed spike, and studies the characteristic representation method and classification strategy suitable for the recognition of superposed spike. In order to solve the problem of waveform information loss caused by superposition of spike waveform, In this paper, the algorithm of phase space reconstruction is studied, and applied to the classification of superimposed spike potential, and a new strategy of spike potential classification based on phase space reconstruction is proposed. Through the phase space reconstruction of the spike potential, the waveform information can be enriched, and the superposition spike sample can be analyzed from many angles. In this paper, a large number of experiments are carried out on the Waveclus data set. The experimental results show that the strategy of spike potential classification based on phase space reconstruction has a good effect on stack spike classification. The accuracy has been improved to a certain extent. Aiming at the problems of large data dimension and poor anti-noise ability in the classification strategy based on phase space reconstruction, this paper improves it. A new method based on the combination of window slope representation and second-order difference representation is introduced in the phase space. Combined with the method of feature optimization, not only the dimension of spike sample can be reduced, but also the variation trend of spike waveform can be highlighted, which is helpful to improve the accuracy of spike potential classification. The experimental results of spike potential data with different noise levels and different superposition degrees show that compared with phase space reconstruction algorithm and window slope representation alone, The combined feature representation method not only reduces the dimension of the data to a certain extent, but also maintains a high classification accuracy for the noisy spike data.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7;R741

【相似文献】