Research on Heart Rate Signal Recognition based on Improved LSTM Network

ECG signal is an important means for diagnosing arrhythmia diseases. Utilizing artificial intelligence algorithms for ECG signal recognition can greatly improve diagnostic efficiency. In response to the issues of vanishing and exploding gradients in traditional Long Short-Term Memory (LSTM) networks, making it difficult to handle long-term dependencies in sequential data, this paper proposes an improved LSTM network model for arrhythmia recognition. Firstly, the original ECG signals are processed by wavelet transform to increase the diversity of temporal and spectral features of the ECG signals. Then, self-attention mechanism is incorporated into the Bidirectional LSTM network to construct the arrhythmia recognition network model. Experimental results show that the improved BiLSTM network for arrhythmia signal recognition is able to effectively identify five types of ECG signals including Normal beat, Supraventricular ectopic beat, Ventricular ectopic beats, Fusion beats, and Unclassified beats, with a mean Average Precision (mAP) of $\mathbf{9 8 . 1 5 \%}$.