深度稀疏自编码器在ECG特征提取中的应用

针对心电（ECG）信号智能分析模型中，复杂波形的特征提取困难，人工设计特征造成源信号特征丢失，标签样本不足等问题，提出了一种基于深度稀疏自编码器（Deep Sparse Auto-Encoders，DSAEs）的ECG特征提取方法。该方法在DSAEs进行贪婪逐层训练时，采用适应性矩阵估计（Adaptive moment estimation，Adam）对网络权重进行寻优，以此获得最优参数组合，同时提取出高层隐含层的输出，并作为ECG高度抽象的低维特征。最后利用支持向量机（Support Vector Machines，SVM）构建分类模型，完成对ECG的特征分类。使用MIT-BIH心律失常数据库的ECG数据进行仿真实验，结果表明，提出的ECG特征提取方法能有效地分层抽取特征，提高分类识别准确率。

Application of Deep Sparse Auto-Encoders in ECG Feature Extraction

In view of the difficulties in feature extraction of complex waveforms in intelligent analysis model of ECG signals, the loss of source signal features caused by artificial design features, and the shortage of label samples, an ECG feature extraction method based on Deep Sparse Auto-Encoders（DSAEs） is proposed. In the greedy layer-by-layer training of DSAEs, this method uses Adaptive moment estimation（Adam） to optimize the network weights to obtain the optimal parameter combination, and simultaneously extracts the output of the higher-level hidden layer as the ECG’s highly abstract low-dimensional features. Finally, Support Vector Machines（SVM） is used to construct a classification model to complete the ECG feature classification. Using the ECG data of MIT-BIH arrhythmia database for simulation experiments, the results show that the ECG feature extraction method proposed in this paper can effectively extract features hierarchically and improve the accuracy of classification recognition.