基于自动睡眠分期的多模态残差时空融合模型  

A Multimodal Residual Spatial-temporal Fusion Model Based on Automatic Sleep Classification

作　　者：郭业才[1,2] 仝爽 Guo Yecai;Tong Shuang(School of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;School of Electronics and Information Engineering,Wuxi University,Wuxi 214105,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]无锡学院电子信息工程学院,江苏无锡214105

出　　处：《系统仿真学报》2024年第9期2065-2074,共10页Journal of System Simulation

基　　金：国家自然科学基金(61673222)。

摘　　要：高精度的睡眠分期对于正确评定睡眠情况起到了至关重要的作用。针对现有的卷积网络无法获取生理信号拓扑特征的问题,提出了一种基于多模态残差时空融合的睡眠分期算法。利用短时傅里叶变换和自适应图卷积获取时频图像和时空图像,将其转换为高维的特征向量;通过时频特征和时空特征提取模块实现特征信息流的轻量化交互;使用特征增强融合模块融合特征信息,输出睡眠分期结果。结果表明:该模型具有较高的准确率,在ISRUC-S3数据集上整体准确率为85.3%,F1分数为83.8%,Cohen's kappa为81%,N1阶段准确率达到69.81%。ISRUC-S1数据集上的实验证明了模型的普遍性。Highly accurate sleep staging plays a crucial role in correctly assessing sleep conditions.Aiming at the problem that the existing convolutional network cannot obtain the topological characteristics of physiological signals,a sleep staging algorithm based on multi-modal residual spatio-temporal fusion is proposed.Time-frequency images and spatio-temporal images are obtained using short-time Fourier transform and adaptive map convolution,which are converted into high-dimensional feature vectors;lightweight interaction of feature information flow is realized through time-frequency feature and spatiotemporal feature extraction modules;the feature enhancement fusion module fuses feature information to outputs sleep staging results.The results show that the model has a high accuracy.On the ISRUC-S3 data set,the overall accuracy is 85.3%,the F1 score is 83.8%,Cohen’s kappa is 81%,and the N1 stage accuracy reaches 69.81%.Experiments on the ISRUC-S1 dataset demonstrate the generality of the model.

关 键 词：睡眠分期 多视图融合 图卷积网络 深度学习 脑电信号 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]