基于多指标融合小波包分解与SVM的脑磁信号分类

Magnetic Signal Classification Based Multiple Indicator Fusion Wavelet Packet Decomposition and SVM

脑磁信号作为一种新的脑机接口输入信号，由于其信号微弱，背景噪声强，是一种随机性很强的非平稳信号．所以在引进了小波包分解基础上，利用多指标融合方法确定最佳分解尺度后，把该尺度下的分解系数作为脑磁信号的特征向量，并利用三种不同核函数的支持向量机对该特征向量进行学习选出最佳参数，然后对含有手运动方向模式信息的脑磁信号进行测试，并与其他5种算法进行比较，其平均分类正确率低于PCA与LDA融合的方法和改进半监督聚类方法，但超过了EMD与AR融合的方法，EMD与Hillbert变换融合的方法以及MVAR与MPCA融合的方法．实验表明了算法在非高斯、含噪声的脑磁信号分类是有效．

As a new brain-computer interface input signal, brain magnetic signal is a highly random non-stationary signal because of its weak signal and strong background noise.In this paper,based on the introduction of wavelet packet decomposition, the use of multi-index fu- sion method to determine the best decomposition scale, The wavelet packet decomposition coefficient at this scale is taken as the eigenvector of the brain magnetic signal, And the sup- port vector machine of three different kernel functions is used to select the best parameters, Then the brain motion signal containing the direction of hand movement direction is tested and compared with the other five algorithms~ The average classification rate is lower than that of PCA and LDA, and the improved semi-supervised clustering method, but more than EMD and AR methods, EMD and Hillbert methods as well as MVAR and MPCA methods. Experiments show that the algorithm is effective in non - Gaussian, noise - free brain magnetic signal classification.