基于多类最小二乘支持向量机的神经元信号识别

Cortical Neural Signals Recognition Based on Multi-class Least Squares Support Vector Machines

针对大脑运动皮层群体神经元信号与运动行为关系的分析,提出一种基于二叉树的最小二乘支持向量机多类分类算法。在对猴子进行三维空间中8个方向手臂运动实验记录的多通道神经元信号的分析中,通过与标准支持向量机和学习矢量量化神经网络的比较,说明该方法不仅与标准支持向量机同样具有比学习矢量量化方法更强的学习能力和预测能力,而且运算时间比标准支持向量机更短。比较结果表明最小二乘支持向量机对于神经元信号分析的有效性和优越性,进而有利于实现性能更高的用于神经康复的脑机接口系统。

In this paper, a multi - class least squares support vector machines （LS- SVM） algorithm of a binary tree recognition strategy is used to analyze the motor cortical neural signals. The neural ensemble data were recorded simultaneously with kinematics of arm movement while the monkey performed reaching tasks from the center position to eight peripheral targets in a three - dimensional virtual environment. The performance of the LS- SVM based neural activity recognition was compared with that of the standard SVM and the learning vector quantization （LVQ） approach. The results show that the LS- SVM and SVM have smaller empirical risks and better generalization performance than the LVQ approach, hut also LS- SVM costs less computational time than SVM. That demonstrates the LS-SVM algorithm is a suitable approach for brain neural signals analyses, and LS- SVM method holds hope for a possibly more accurate brain - computer interface for neural prosthesis.