基于卷积神经网络的非接触式呼吸暂停算法研究

Research on non-contact apnea algorithm based on Convolutional Neural Network

本文基于压电陶瓷传感器信号,提出了一种一维卷积神经网络的睡眠呼吸暂停综合征检测算法.该算法利用嵌入智能床垫的压电陶瓷传感器采集头部运动作为输入信号.卷积神经网络模型包括6层的卷积层,每层包含一个ReLU激活函数,一个批归一化(Batch Normalization,BN)层、一个dropout层以及一个最大池化层.同步采集了11位测试者的压电陶瓷传感器信号和多导睡眠图信号,生成了40988个样本,正负样本均衡.训练集、验证集、测试集按照60%、20%、20%的比例进行划分.最终,本文所提出的检测模型在测试集上得到了92.76%的准确率,88.67%的精准率,98.06%的召回率,93.13%的F1-得分.

Based on the piezoelectric ceramics sensor signal,this paper proposes a one-dimensional Convolutional Neural Network detection algorithm for sleep apnea syndrome.The Convolutional Neural Network model consists of six layers of convolutional layers.Each layer containing a ReLU activation function,a batch normalization layer,a dropout layer,and a max pooling layer.Simultaneously,the piezoelectric ceramics sensor signals and polysomnographic signals of 11 subjects are collected and 40 988 balanced samples are generated.The training set,verification set and test set are divided according to the ratio of 60%,20%and 20%.Based on the above,using the model proposed in this paper,an accuracy of 92.76%,a precision of 88.67%,a recall of 98.06%and an F1-score of 93.13%are attached with the test dataset.