应用生长、分级的自组织映射模型进行意识任务分类

GROWING HIERARCHICAL SELF-ORGANIZING MAP MODELS FOR MENTAL TASK CLASSIFICATION

提出一种使用生长、分级的自组织映射(growinghierarchicalself-organizingmap,GHSOM)模型进行基于EEG信号的意识任务分类来实现脑机接口技术的方法。GHSOM模型是自组织映射(self-organizingmap,SOM)的一种变体,由多层的SOM组成,具有一定的分级结构,能够表达数据中不同层次的信息。同时研究了使用平均量化误差(meanquantizationerror,mqe)和量化误差(quantizationerror,qe)两种方法实现的GHSOM模型对意识任务分类的作用。结果表明,GHSOM模型对于意识任务的可分性能够提供可视化的信息,并且发现使用量化误差方法实现的GHSOM模型提供较多的数据信息和较高的分类精度。使用GHSOM模型进行了5类意识任务的分类,平均分类精度可达80%。

The growing hierarchical self-organizing map （GHSOM） model was proposed to apply to performing mental tasks classification in EEG for Brain-Computer Interface. The GHSOM model is a variant of SOM, and consists of many layers of SOMs, which form the hierarchical architecture. The hierarchical structure hid in data can be expressed by GHSOM model. The effectiveness of GHSOM models implemented using both the mean quantization error （mqe） and quantization error （qe） methods for mental tasks classification was investigated. The results indicated that GHSOM models provided visual information about the separability of different mental tasks, and the GHSOM model using quantization error method provided more detailed information about data and obtained high classification accuracy. About 80% of the average classification accuracy for five mental tasks classifications was achieved by using the GHSOM model.