基于听觉诱发脑-机接口的识别模型研究  

Research on Recognition Model Based on Auditory Evoked Brain-Computer Interface

作　　者：魏佳鑫 张晓飞 龚真颖 郭一娜[1] WEI Jia-xin;ZHANG Xiao-fei;GONG Zhen-ying;GUO Yi-na(College of Electronic Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学电子信息工程学院,太原030024

出　　处：《太原科技大学学报》2022年第6期489-495,共7页Journal of Taiyuan University of Science and Technology

基　　金：山西省重点研发计划(201803D421035)。

摘　　要：针对听觉诱发脑-机接口存在的“BCI盲”、泛化能力差等问题,设计基于听觉诱发的脑-机接口实验范式,提出了惩罚式长短期记忆神经网络融合全连接层的识别算法。首先,将实验采集到的脑电数据处理后作为神经网络的数据集输入,然后对长短期记忆神经网络中输出门的损失函数添加惩罚项,减少模型的参数,将其输出输入到DENSE层,解决模型训练过程中不易收敛的问题。实验表明,文中算法的识别率达到91.59%,解决了“BCI盲”的问题,有效解决了算法过拟合与不易收敛的问题。其分类性能不仅高于长短期记忆神经网络,而且相比一些其他代表性的算法也有一定优势。Aiming at the problems of BCI blindness and poor generalization ability of the Brain-induced Brain Computer Interface, a brain-computer interface experimental paradigm based on auditory evoked is designed, and a recognition algorithm of Regularizer Long Short-Term Memory neural network and fully connected network is proposed.First, the EEG data collected by the experiment is processed as a neural network data set input, and then a penalty term is added to the loss function of the output gate in the long and short-term memory neural network to reduce the parameters of the model and input its output to the DENSE network.The problem of difficulty in convergence during model training is solved.Experiments show that the recognition rate of the algorithm reaches 91.59%,which solves the problem of BCI blindness and effectively solves the problem of overfitting and difficult convergence of the algorithm.Its classification performance is not only higher than that of long-term and short-term memory neural networks, but also has certain advantages over some other representative algorithms.

关 键 词：脑-机接口 听觉诱发 长短期记忆神经网络 L2范数正则化 全连接层 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]