重庆理工大学学报2025，Vol.39Issue(19)：141-148,8.DOI:10.3969/j.issn.1674-8425(z).2025.10.017
融合动态重加权协同训练的半监督睡眠分期方法
Research on semisupervised sleep staging method with integrated dynamic re-weighting co-training
摘要
Abstract
To address the pseudo-label noise and model convergence in semi-supervised sleep staging,this paper proposes a dynamic re-weighting co-training(DRCT)framework.It enhances model robustness and pseudo-label reliability through synergistic optimization and dynamic sample re-weighting.During the pre-training,subset partitioning or construction of different model architectures enhance initial model diversity.During retraining,the framework implements cross-model pseudo-label interactions and introduces a consistency-based dynamic re-weighting mechanism.The mechanism prevents undesirable model convergence and optimizes classification performance.Experimental results demonstrate it achieves an accuracy of 81.2%and 79.8%on the Sleep-EDF and Sleep-EDFx datasets,significantly outperforming existing approaches.关键词
动态重加权/协同训练/半监督/睡眠分期/伪标签Key words
dynamic re-weighting/co-training/semi-supervised/sleep staging/pseudo-label分类
计算机与自动化引用本文复制引用
李华,赵文丽,张航,李奇,武岩,刘方姿..融合动态重加权协同训练的半监督睡眠分期方法[J].重庆理工大学学报,2025,39(19):141-148,8.基金项目
吉林省科技发展计划项目(20240101344JC,20230203098SF) （20240101344JC,20230203098SF）
中山市社会福利与基础研究项目(2023B2015) （2023B2015）
