基于小波包最优基的运动想象EEG自适应特征提取方法

Adaptive Feature Extraction Based on Best Basis of Wavelet Packet for Motor Imagery EEG

针对运动想象脑机接口系统存在分类正确率低、自适应能力差等不足,提出一种基于小波包最优基的自适应特征提取方法;该方法首先对运动想象EEG进行小波包分解;其次,对传统的距离准则进行改进,通过引入权重因子表征对类内距离和类间距离的关注程度,获得一种既可满足小波包最优基评价准则的可加性条件,又有效地增强了频带特征信息的可分离性的评价准则;进而,采用"自底向顶、自左至右"的快速搜索策略获取小波包最优基,并选取最优基对应的分类性能评价值较高的部分频带小波包系数构成分类特征;仿真结果表明本方法最高分类正确率可达93.4%,与常用的时频分析方法对比,验证了本算法具有较高的分类正确率和较小的时间花费。

In brain--computer interfaces of imagery movement, a new method which can adaptively extract features on the basis of the best wavelet package basis is proposed to solve the problems such as the low classification accuracy and weak self--adaptation, etc. First, wavelet packet is exploited to decompose the motor imagery EEG signal; second, the traditional distance criterion is optimized by introducing a weight parameter which reflects the importance of including both the interelass and the intraclass inertias, and this evaluation criterion for classification is not only under the condition that the criteria is additive for the choice of the best wavelet packet basis, but also can effectively improve the separability of the feature information in frequency subbands; Third, the best wavelet packet basis is attained by using a fast search strategy of ＂from the bottom to the top. from the left to the right＂, and the classification feature is extracted by choosing the part wavelet package coefficient which can attain higher value by calculating the classification evaluation criterion according to the best 4oasis; The experimental results which are compared with another common method of time and frequency analysis, show that the algorithm could produce high classification accuracy and less time consumption.