基于Fisher+Fuzzy算法提高SSVEP脑电信号分类

Based on Hybrid Fisher and Fuzzy Algorithms to Improve Classification Accuracy of EEG-Based SSVEP Brain Signals

为了提高脑-机接口(Brain-Computer Interface,BCI)系统中基于稳态视觉诱发电位(Steady-state Visual Evoked Potentials,SSVEP)信号的分类准确率,提出了一种新的基于Fisher+Fuzzy的分类算法。该算法首先对提取的脑电特征利用Fisher算法得到最佳投影方向和阈值,然后对样本点到最佳超投影面的距离d进行模糊化,再通过模糊推理确定分类结果。该分类算法改善了在SSVEP分类中使用单一Fisher分类器难以对多分类问题中处于歧义区的样本进行有效分类的问题。结果显示在SSVEP的三、四、五分类中,Fisher+Fuzzy分类器取得了94.72%,92.18%,86.08%的平均分类准确率,高于单一Fisher分类器90.07%,80.60%,74.42%的平均准确率,对具有较低可分性的数据集进行分类时准确率显著提高。

In order to improve the classification accuracy of electroencephalographic based on the steady-state visual evoked potential(SSVEP) in brain computer interface(BCI), a new classification algorithm combining Fisher and Fuzzy is proposed in this paper. First, the algorithm uses Fisher to obtain the optimal projection direction and the threshold value for the EEG signals. Second, calculate the distance d and fuzzy it. Finally, the classification result is obtained by fuzzification and defuzzification process. The classification algorithm overcomes the shortcoming that the samples in the ambiguous area cannot be accurately classified by using a single Fisher classifier in SSVEP for multiple classification problems. In the three, four and five-classification based on the SSVEP, the classification algorithm proposed in this paper has achieved 94.72 %, 92.18 % and86.08% average classification accuracy that are higher than using a single Fisher classifier achieved 90.07 %,80.60% and 74.42%. Faced with the low separability data set, the algorithm can significantly improve the classification accuracy.