摘要
为提高基于心电信号的睡眠呼吸暂停检测精度,针对现有检测方法普遍存在的需要较复杂的特征工程和手工校正步骤,无法对心电信号自适应预处理且损失较多信息的问题,提出一种基于空洞卷积和注意力机制的睡眠呼吸暂停检测方法。首先,利用自适应预处理网络滤除心电信号中的冗余信息(包括基线漂移、肌电干扰等);然后,使用基于空洞卷积和时间注意力机制的检测网络从心电信号中提取时序特征并进行检测。在Apnea-ECG数据集上的实验结果表明,相比其他6种检测方法,本文方法能够实现更有效的睡眠呼吸暂停检测。
In order to improve the accuracy of sleep apnea detection based on ECG signal,aiming at the problem that the existing detection methods generally need more complex feature engineering and manual correction steps,can not adaptively preprocess ECG signal and will lose more information,a sleep apnea detection method based on dilated convolution and attention mechanism is proposed.Firstly,the adaptive preprocessing network is used to filter the redundant information in ECG signal(including baseline drift,EMG interference,etc.);Then,the detection network based on dilated convolution and time attention mechanism is used to extract timing features from ECG signal and detect them.The experimental results on Apnea-ECG data set show that compared with the existing detection methods,this method can achieve more effective sleep apnea detection.
作者
郑和裕
林美娜
ZHENG Heyu;LIN Meina(Guangdong University of Technology,Guangzhou 510006,China)
出处
《自动化与信息工程》
2022年第2期29-34,40,共7页
Automation & Information Engineering