一种深度学习模型的研究与应用

Research and Application of a Deep Learning Model

深度学习作为近年来快速发展的崭新技术可以有效帮助研究目标检测和模式识别,在信号与信息处理领域成为研究热点。针对胎儿心电信号难以提取导致胎心率检测困难,设计了一种深度学习模型。该模型使用了卷积神经网络结构,并且结合了批量标准化和Dropout技术,可以在不去除母体心电信号的情况下直接检测胎儿QRS波群。该方法首先在PhysioNet上选取母体腹部心电信号作为实验数据集,然后通过样本熵进行信号质量评估,预处理去除电力线干扰和基线漂移干扰,最后分段进行短时傅里叶变换将一维心电信号转化为二维时频图,再通过卷积神经网络进行分类。实验结果表明,该方法可以取得较高的灵敏度(86.98%)、阳性预测值(88.35%)和准确率(78.03%)。通过对比支持向量机和BP神经网络两种算法在相同数据集上的准确率,验证了卷积神经网络在分类性能上更具有优势。

Deep learning,as a new technology developed rapidly in recent years,can effectively help research target detection and pattern recognition,and has become a research hotspot in the field of signal and information processing. A deep learning model is designed to detect fetal heart rate due to the difficulty in extracting fetal ECG signals. With the convolutional neural network structure,combined with batch normalization and dropout technology,the model can be able to directly detect fetal QRS complexes without removing maternal ECG signals. In this method,maternal abdominal ECG signals are first selected as the experimental data set on the PhysioNet,then the sample entropy method is used for signal quality assessment,and power line interference and baseline drift interference are removed by preprocessing. Finally,one-dimensional ECG signals can be converted to two-dimensional time-frequency diagrams through short-time Fourier transform by segmenting,and then the classification is carried out by convolutional neural network. Experiment shows that the proposed method can achieve higher sensitivity(86.98%),positive predictive value(88.35%) and accuracy(78.03%). By comparing the accuracy of SVM and BP neural network on the same data set,it is verified that convolutional neural network has more advantages in classification performance.