基于卷积神经网络的不同病理的鼾声分类

Classification for Different Pathologies Snoring Sounds Based on Convolutional Neural Network

本文提出用卷积神经网络对阻塞性呼吸暂停和中枢性呼吸暂停所致的鼾声进行分类,并提出了鼾声的完整上气道冲激响应特征,设计了1维卷积神经网络对这两种鼾声做了分类处理和识别.运用多组输入特征对该网络分类性能做了评估,结果表明完整上气道冲激响应特征作为输入参数的平均正确率达到79%,高于其他特征作为输入参数的结果.

In this paper,we present to use convolutional neural network to classify two different pathological snoring sounds,these two pathologies snoring are caused by obstructive sleep apnea and central sleep apnea respectively during night.We consider snoring caused by different pathologies with different acoustic characteristics in signal processing,propose a complete upper airway impulse response feature of snoring,and use our designed one-dimensional convolutional neural network to classify and recognize two type snoring sounds.Multiple sets of input features are used to evaluate the performance of the ID CNN classification.The results show that the average accuracy of the complete upper airway impulse response feature as an input parameter reaches 79%,which is higher than the results of other features as input parameters.