半监督元学习在意识障碍患者情绪识别中的应用  

Application of Semi-supervised Meta-learning on the Emotion Recognition of Patients with Disorder of Consciousness

作　　者：李景聪 陈希昶 黄海云 LI Jingcong;CHEN Xichang;HUANG Haiyun(School of Software,South China Normal University,Foshan 528225,China)

机构地区：[1]华南师范大学软件学院,广东佛山528225

出　　处：《软件导刊》2025年第4期12-17,共6页Software Guide

基　　金：国家自然科学基金项目(62306120)。

摘　　要：情绪识别是脑机接口一个主流的应用方向,有着巨大的研究潜力。然而,被试之间存在的生理差异会使脑电信号差异较大,并且意识障碍(DOC)患者的有标签情绪脑电数据极为稀缺。为此,提出一个利用少量有标签数据和大量无标签数据来实现跨被试情绪识别任务的模型,并结合半监督与元学习方法微调模型,以提升数据利用率。在自采数据集中的实验表明,所提模型识别正常人、DOC患者情绪的准确率分别为82.93%、59.26%,证实了该方法在跨被试情绪识别任务和DOC患者诊断中的潜力巨大。Emotion recognition is a mainstream application direction of brain-computer interface,which has great research potential.Howev‐er,the physiological differences between participants can lead to large differences in EEG signals,and labeled emotional EEG data in patients with impaired consciousness(DOC)are extremely scarce.Therefore,a model using a small amount of labeled data and a large amount of unla‐beled data to realize the cross-participant emotion recognition task was proposed,and the model was fine-tuned by combining semi-super‐vised and meta-learning methods to improve data utilization.Experiments in the self-collected dataset show that the accuracy of the proposed model in identifying emotions in normal people and DOC patients is 82.93%and 59.26%,respectively,which confirms that the method has great potential in the cross-participant emotion recognition task and the diagnosis of DOC patients.

关 键 词：情绪识别 脑机接口 跨被试 脑电图 元学习 半监督 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]