基于共同空间模式的扫视信号特征提取算法

Research on feature parameters extracting algorithm of saccade signals based on common spatial pattern

为了提高基于眼电图(EOG)的扫视信号识别正确率,提出了一种基于共同空间模式(CSP)的扫视信号特征提取算法.该算法首先利用事先标注好的标签数据进行CSP空域滤波器设计,并采用联合近似对角化的方法解决多分类问题;在此基础上,使用该滤波器对原始多导联眼动信号进行空域滤波,滤波输出即为扫视信号的特征参数.在实验室环境中使用支持向量机对上、下、左、右四类扫视信号进行识别,所提算法的平均正确率达到了97.7%.实验结果表明基于CSP的扫视信号特征提取算法在眼动信号分析中呈现出良好的分类性能.

In order to improve the recognition accuracy of saccade EOG （electro-oculogram） signals, a feature parameters extracting algorithm of saccade EOG signals based on common spatial pattern （CSP） was proposed. The labeled data were used to train different CSP spatial filters, and the joint approximate diagonalization method was utilized to solve multi-classification question. Then, original multi-channel EOG signals were filtered by the established CSP spatial filter. Finally, the output re- suits were regarded as feature parameters of saccade EOG signals in different eye movements＇ tasks. In lab environment, the support vector machine （SVM） was adopted to classify four type＇s saccade signals, i. e. , up, down, left and right）. Experiential results show that the recognition accuracy is about 97.7%, which reveals that the proposed algorithm has a good classification performance in saccade EOG signals analysis.