基于可生长自组织映射的意识任务分类

在基于脑电（EEG）的脑一机接口技术中，使用可生长自组织映射（SOM）神经网络进行了5类意识任务分类的研究．结果表明：①可生长SOM能够根据数据内部结构自适应地调整确定其映射网络的拓扑形状，在一定程度上反应了数据的分布特征；②可生长SOM更关注那些表达误差比较大的映射单元，从而整体上减小了映射网络的表达误差，提高了对数据模式的表达能力，有利于模式的分类处理；③可生长SOM侧重于表达类别之间的边界信息，这对于分类问题有着积极的作用．与传统SOM相比，使用可生长SOM进行5类分类处理得到的分类精度提高了10％左右，分类正确率可以超过80％，说明可生长SOM在脑-机接口系统中有着很大的潜在应用性．

Mental Tasks Classification Based on Growing Self-Organizing Map

The growing self-organizing map(Growing SOM) was adopted to perform mental tasks classification in EEG-based brain-computer interface(BCI).The Growing SOM is endowed with some capacities to outperform: Adapting the topological shape of the mapping network according to the inherent structures of the input data to reflect the features of the data distribution;Focusing on those mapping units with greater expression errors to decrease the whole expression error of the mapping network to result in the more accurate representation of the input data;Emphasizing the expression of the regions between different classes of data to obviously improve the classification performances.Compared with the traditional SOM,the Growing SOM with an accuracy of higher than 80% in mental tasks classification demonstrates the great potential applicability in BCI systems.