虚拟人引导的脑电信号重要性选择与识别.

In order to maintain better recognition robustness in the EEG signals with noise data or missing value, and to reveal the interaction relationship between multiple channels of EEG signals, a series of studies have been carried out. This paper used the random forest algorithm to select the important channels with interaction and remove the irrelevant and redundant channels. It used the state space model to describe the internal motion law of multi-channel, reflecting the relationship between internal state and input or output. It used the EM algorithm to identify the parameters, as the identification features of the state space model. Moreover, this paper identified the extracted features by SE-GRU model, which increased the weight of important features. Generally, this method can effectively improve the classification accuracy on both the public datasets and virtual human guided EEG datasets. Compared with the methods without channel selection, it achieves better results. Meanwhile, through using the final trained model, it achieves the control of virtual human. [ABSTRACT FROM AUTHOR]