论文部分内容阅读
卷积神经网络已成为当前图像识别和语音分析领域的研究热点,并在图像分类等二维信号的问题处理中取得了良好效果。本文成功地将卷积神经网络用于解决一维心电信号的有关问题中。RR间期绝对不规则和P波消失,代之以连续不等的f波是发生房颤时的两个重要心电图表现。而RR间期不规则亦能体现在其他心律失常之中,同时P波或f波属微弱信号其特征点检测较为困难且形状特征难以把握。因此本文提出了一种基于单心拍心房活动特征与卷积神经网络房颤检测方法。首先把所有心电信号归一化处理为长度相等的单心拍,然后对所有心拍进行白化、求解每类心拍的稀疏系数、对所求稀疏系数进行池化处理,最后使用卷积神经网络进行心电信号分类达到房颤检测的目的实验结果表明该方法检测结果的正确率为95.91%,为检测房颤提供了很好的选择。
Convolutional neural networks have become the hotspot in the field of image recognition and speech analysis, and have achieved good results in the processing of two-dimensional signals such as image classification. In this paper, we have successfully used convolutional neural networks to solve the problems related to one-dimensional ECG signals. RR interval is absolutely irregular and P wave disappears, replaced by a continuous range of f-wave is the occurrence of two major ECG occurs when the ECG. The RR interval irregularity can also be reflected in other arrhythmias, while P wave or f wave is a weak signal detection of its characteristic points more difficult and difficult to grasp the shape and characteristics. Therefore, this paper presents a heart beat based on single heart activity characteristics and convolutional neural network detection of atrial fibrillation. First, all ECG signals are normalized to a single heartbeat with the same length, and all the heartbeats are whitened. The sparse coefficients of each type of heartbeat are solved, and the sparse coefficient is spatially pooled. Finally, a convolutional neural network is used to perform heartbeat The purpose of the electrical signal classification to atrial fibrillation test The experimental results show that the detection accuracy of the method is 95.91%, which provides a good choice for the detection of atrial fibrillation.