软件导刊2025，Vol.24Issue(5)：122-129,8.DOI:10.11907/rjdk.241025
基于多尺度时间残差收缩网络的睡眠分期模型
Sleep Staging Model Based on Multi-Scale Temporal Residual Shrinkage Network
摘要
Abstract
Sleep stage classification is an important step in evaluating sleep quality,and automatic sleep staging technology is of great signifi-cance in fields such as sleep monitoring,diagnosis and treatment of sleep disorders,and sleep science research.Deep learning has been prov-en to be an effective method for automatically classifying sleep stages,but there is still room for improvement in performance.Therefore,a sleep staging model based on multi-scale time residual shrinkage network is proposed.Firstly,the sleep stage is classified using a Multi Scale Time Residual Shrinkage Network(MS-TRSN).Two branch convolutional layers automatically extract features of different scales from the 30 second sleep electroencephalogram(EEG)signal using kernels of different sizes,and the Attention Feature Fusion(AFF)module based on attention mechanism fuses features of different scales;Secondly,the Time Residual Shrinkage Module(TRSM)is utilized to capture temporal contextual information and filter out irrelevant features.The results on the Sleep-EDF-20 and Sleep-EDF-78 datasets showed that the accura-cy,macro F1 score,and Cohen's kappa of the proposed model were 86.36%,80.94%,0.81,and 82.39%,77.42%,and 0.76,respectively,demonstrating excellent performance.关键词
自动睡眠分期/深度学习/注意力机制/时间残差收缩Key words
automated sleep stage classification/deep learning/attention mechanism/time residual shrinkage分类
信息技术与安全科学引用本文复制引用
李小鱼,刘雨婷,郜东瑞,汪曼青..基于多尺度时间残差收缩网络的睡眠分期模型[J].软件导刊,2025,24(5):122-129,8.基金项目
四川省科技厅重点研发计划项目(2023YFG0018) （2023YFG0018）
成都信息工程大学科技创新能力提升计划项目(KYQN202241) （KYQN202241）
