【摘要】 心电信号分类是医疗保健领域的重要研究内容.针对大多数方法不能很好地降低样本数量少的类别漏诊率，以及降低预处理操作的复杂性问题，提出了一种基于改进深度残差收缩网络(IDRSN)的心电信号分类算法(即DRSL算法).首先，使用合成少数类过采样技术(SMOTE)扩充数量少的类别样本，从而解决了类不平衡问题；其次，利用改进深度残差收缩网络提取空间特征，其残差模块可以避免网络层加深造成的过拟合，压缩激励和软阈值化子网络可以提取重要局部特征并自动去除噪声；然后，通过长短期记忆网络(LSTM)提取时间特征；最后，利用全连接网络输出分类结果.在MIT-BIH心律失常数据集上的实验结果表明，该算法的分类性能优于IDRSN、DRSN、GAN+2DCNN、CNN+LSTM＿ATTENTION、SE-CNN-LSTM分类算法.更多还原

【Abstract】 The electrocardiogram(ECG) signal classification is a significant research topic in the healthcare field. Most existing methods could not effectively reduce the missed diagnosis rate of classification with small-size samples and tackle the complexity of preprocessing operations. An electrocardiogram signal classification algorithm based on the improved deep residual shrinkage networks was proposed, namely the DRSL algorithm. Firstly, the small-size classification samples were augmented with the synthetic minority over-sampling technique to solve the classification imbalance problem. Secondly, the spatial features were extracted by mean of the improved deep residual shrinkage networks, where the residual module can avoid overfitting caused by deepening of network layers, and the squeeze-and-excitation operation with soft threshold subnetwork can extract important local features and remove noises automatically. Then, the time features were extracted with the long short-term memory networks. Finally, the classification results were output with the fully connected neural networks. The experimental results on the MIT-BIH arrhythmia database show that, the proposed algorithm is superior to IDRSN, DRSN, GAN+2DCNN, CNN+LSTM＿ATTENTION, SE-CNN-LSTM in terms of classification performances.更多还原