Recognition of ECG signals by convolutional neural network based on attentional mechanism

In this paper, a deep network model based on attentional mechanism is proposed to realize ECG recognition. The model is based on convolutional neural network, which is composed of four convolutional layers and full connection layers. In order to improve the performance of feature extraction by convolutional neural network, an attention mechanism module is embedded behind the pooling layer of the first convolutional neural network. The feature maps from output by the pooling layer are sent to the channel domain module and the spatial domain module respectively to generate attention maps. The attention maps assign different weights to the feature maps to complete the re-extraction of ECG signal features and realize adaptive feature refinement. The characteristics of ECG signal were further enhanced through two dimensions. Considering the characteristics of ECG signals and the hard coding characteristics of convolutional neural network calculation, S-transform is used to establish two-dimensional ECG time-frequency matrix images for different time-sequence ECG signals, which is more conducive to signal recognition. Using the MIT/BIH arrhythmia database set of ECG signals, six different ECG signals were identified and classified, this model was verified 5-fold cross validation on the testing set, recognition accuracy is above 99.59%.