基于深度学习的FFT-GAM鼾声音频信号分类方法研究  

Research on Deep Learning-based FFT-GAM Snoring Audio Signal Classification Method

作　　者：富家国 华才健 FU Jiaguo;HUA Caijian(School of Computer Science and Engineering,Sichuan University of Science and Engineering,Yibin 644000,China)

机构地区：[1]四川轻化工大学计算机科学与工程学院,四川宜宾644000

出　　处：《成都工业学院学报》2025年第2期48-57,共10页Journal of Chengdu Technological University

摘　　要：阻塞性睡眠呼吸暂停低通气综合征(OSAHS)是一种由于上呼吸道阻塞引起的慢性呼吸障碍,在患有呼吸障碍的人群中,OSAHS患者占比超过75%。多导睡眠图(PSG)被广泛认为是临床诊断OSAHS患者的金标准,但是需要具备专业设备和专业人员,并且具有侵入性,无法广泛应用。因此,提出一种方便、高精度的OSAHS筛选方法:基于深度学习的快速傅里叶变换的全局注意力机制(FFT-GAM)的网络模型。对武汉大学鼾声数据集(SSBPR)中的数据进行预处理和特征提取,将提取的特征转换为224×224的频谱图,输入到FFT-GAM模型中处理,对OSAHS患者不同睡姿的鼾声进行分类,预测患者睡眠打鼾时的睡姿。实验结果显示,该方法的分类准确率最高达到89.8%,召回率最高提升至91.3%。鼾声与睡眠姿势之间存在显著的关联性,该方法在分类任务中表现优异,具有较高应用潜力。The obstructive sleep apnea hypopnea syndrome(OSAHS)is a chronic respiratory disorder caused by obstruction of the upper respiratory tract,accounting for more than 75%of the population suffering from respiratory disorders.Polysomnography(PSG)is widely regarded as the gold standard for diagnosing OSAHS,but its invasive nature and reliance on specialized equipment and trained professionals,making it difficult to be widely applied.Therefore,a convenient and high-precision screening method for OSAHS was proposed,namely the network model of global attention mechanism(FFT-GAM)based on fast Fourier transform of deep learning.The data in Wuhan university snore-based sleep body position recognition(SSBPR)dataset were pre-processed and feature extraction was carried out,and the extracted features were converted into 224×224 spectral graphs,which were input into FFT-GAM model for processing,so as to classify snoring in different sleeping positions of OSAHS patients and predict the sleeping positions during snoring.Experimental results show that the classification accuracy of this method is up to 89.8%,and the recall rate is up to 91.3%,indicating a significant correlation between snoring and sleep posture.This method performs well in classification tasks and has high application potential.

关 键 词：阻塞性睡眠呼吸暂停低通气综合征 鼾声 睡眠姿势 注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]