结合注意力机制的CNN-LSTM心电信号识别

ECG SIGNAL RECOGNITION BASED ON CNN-LSTM CONBINED WITH ATTENTION MECHANISM

心电信号形态复杂多样易导致识别准确率低、适应性差,通常依靠人工诊断,费时费力。为此提出注意力机制与卷积长短时记忆网络(CNN-LSTM)相结合的深度网络模型(Attention-Based CNN-LSTM,A-CNN-LSTM)以实现心电信号自动识别。模型以CNN为基础架构,引入了注意力机制帮助心电信号内空间特征的提取;LSTM捕捉空间特征内的时间特性,并将其用于信号分类。在MIT-BIH心律不齐数据库上进行实验,结果表明,该模型可对六种不同的心电信号进行分类,识别准确率达到99.23%,具有一定的临床应用意义。

The complexity and diversity of ECG signal forms easily lead to low recognition accuracy and poor adaptability.It's time-consuming,laborious and costly to rely on ECG experts to participate in feature recognition.Therefore,a new deep network model,the A-CNN-LSTM,is proposed that combined with the attention mechanism,long short-term memory and convolutional neural network.This model was built on convolutional neural network,where the attention mechanism was introduced after the pooling layer of the convolutional neural network to help extracting spatial features of the ECG signal.The temporal feature within them could be captured by LSTM to be used in classification.The experiment was conducted on the MIT-BIH arrhythmia database.The experimental results show that this model can classify six different kinds of ECG signals and have achieved a classification accuracy of 99.23%.The model is of certain application significance clinically.