基于局部场电位的动物转向解码研究

发布时间：2018-08-28 11:36

【摘要】：运动行为的神经信号解码研究是脑-机接口研究的重要内容,解码生物的运动意图,具有重要的理论与实际应用价值。微电极阵列记录到的信号包括锋电位信号(spike)和局部场电位信号(Local field potentials,LFP),关于spike信号的运动解码研究取得了一定成果,但是随着电极植入的时间增长,spike信号质量有所下降,而局部场电位具有长期解码稳定的特点,逐渐引起研究人员的关注。但是,大脑结构复杂,某一运动行为的执行可能由大脑多个脑区共同作用,运动意图在脑中的存在时间非常短暂,采集到的局部场电位为非平稳信号,且在记录的过程中不可避免的会引入噪声信号,这些都对解码特征的有效提取造成一定阻碍。在对大脑信息进行解码时,关键是确定出有效编码信息的时间及频率窗口,提取到利于解码的特征。本文以鸽子为研究对象,结合动物行为学方法与神经信号微电极阵列检测技术,采集了鸽子左转、直行、右转三个方向转向运动发生时NCL(Nidopallium Caudolaterale)脑区的LFP信号。对LFP信号特征提取方法进行分析,探讨鸽子运动转向时的LFP信号的特征变化,并用k近邻(k-Nearest Neighbor,kNN)方法对提取的特征进行分类,预测其运动方向。主要工作如下:1.对局部场电位信号的产生过程,局部场电位信号特性及相关噪声特性进行了分析,在此基础上比较了常用处理方法在LFP信号去噪中的优缺点,以及常用时域,频域时频域方法在特征提取方面的优缺点。2.采用了一种结合独立成分分析(Independent component analysis,ICA)与小波方法的特征提取方法,ICA方法能够去除各通道间的数据冗余,去除明显噪声的同时对有效编码信息最大程度的保留,之后利用时频特性特较好的小波方法进行特征提取,有效提取了运动转向发生时局部场电位的特征。3.完成了鸽子运动转向行为诱导训练及神经信号采集实验,利用本文的特征提取方法对局部场电位信号进行特征提取,并利用k近邻方法进行分类,正确率最高达到92.35%。又进一步对不同通道,单个特征,及不同时间窗提取的特征进行分类并统计正确率,探究比较好的解码特征形式。
[Abstract]:The study of neural signal decoding of motor behavior is an important part of brain-computer interface research. Decoding biological motion intention has important theoretical and practical application value. The signals recorded by microelectrode array include spike signal (spike) and local field potential signal (Local field potentials,LFP). Some achievements have been made in the study of motion decoding of spike signal, but the quality of spike signal has declined with the time of electrode implantation. The local field potential has the characteristics of long-term decoding stability, and gradually attracted the attention of researchers. However, the structure of the brain is complex, the execution of a certain motor behavior may be affected by multiple brain regions, the duration of the motor intention in the brain is very short, and the local field potential collected is a non-stationary signal. Noise signals will inevitably be introduced in the process of recording, which will hinder the efficient extraction of decoding features. When decoding the brain information, the key is to determine the time and frequency window of the effective coding information, and extract the features that are beneficial to the decoding. In this paper, pigeons were studied. The LFP signals in NCL (Nidopallium Caudolaterale) brain region were collected when pigeons turned left, straight and right, combined with animal behavior method and neural signal microelectrode array detection technique. The feature extraction method of LFP signal is analyzed, and the feature change of LFP signal when pigeon is moving and turning is discussed. The extracted feature is classified by k-nearest neighbor (k-Nearest Neighbor,kNN) method, and its motion direction is predicted. The main work is as follows: 1. The generation process of local field potential signal, the characteristics of local field potential signal and related noise characteristics are analyzed. On this basis, the advantages and disadvantages of common processing methods in LFP signal denoising are compared, as well as the common time domain. The advantages and disadvantages of time-frequency domain method in feature extraction. A new feature extraction method based on Independent component Analysis (Independent component analysis,ICA) and wavelet method is proposed, which can remove the data redundancy between the channels, remove the obvious noise, and keep the effective coding information to the maximum extent. After that, the feature extraction is carried out by wavelet method, which is especially good in time-frequency characteristic, and the feature of local field potential at the time of motion turn is extracted effectively. 3. The training of pigeon movement steering behavior and the experiment of nerve signal acquisition were completed. The feature extraction method of this paper was used to extract the feature of local field potential signal, and the k-nearest neighbor method was used to classify the local field potential signal. The highest correct rate was 92.35%. Furthermore, the features extracted from different channels, single features and different time windows are classified and the correct rate is counted to explore the better decoding feature form.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN911.7

【相似文献】