陈庆端
（广东工业大学，广东 广州510006）
摘要：睡眠分期对睡眠质量评估、睡眠障碍诊断具有重要的意义。针对基于深度学习的睡眠分期存在标签数据少、数据标注困难等问题，提出一种CLIP自监督学习的多模态睡眠分期方法。通过学习无标签数据的特征表示，解决了因标签数据少而导致的模型训练效果欠佳的问题。在不同标签数据下，将基于CLIP的多模态自监督学习方法与有监督学习、单模态自监督学习方法SimCLR和TS-TCC进行对比实验。实验结果表明，基于CLIP的多模态自监督学习方法能有效提高睡眠分期的性能。
关键词：多模态自监督学习；睡眠分期；CLIP；单模态自监督学习；有监督学习
中图分类号：TN911.7; R318 文献标志码：A 文章编号：1674-2605(2024)04-0004-07
DOI：10.3969/j.issn.1674-2605.2024.04.004 开放获取
A Multi Modal Sleep Staging Method of CLIP Self Supervised Learning
CHEN Qingduan
(Guangdong University of Technology, Guangzhou 510006, China)
Abstract: Sleep staging is of great significance for assessing sleep quality and diagnosing sleep disorders. A multi modal sleep staging method of CLIP self supervised learning is proposed to address the problems of limited labeled data and difficulty in data annotation in deep learning based sleep staging. By learning the feature representation of unlabeled data, the problem of poor model training performance caused by limited labeled data has been solved. Comparative experiments will be conducted between CLIP based multi modal self supervised learning method and supervised learning, single modal self supervised learning methods SimCLR and TS-TCC under different labeled data. The experimental results indicate that the multi modal self supervised learning method based on CLIP can effectively improve the performance of sleep staging.
Keywords: multi modal self supervised learning; sleep staging; CLIP；single modal self supervised learning; supervised learning
