ECG signal classification model based on multi-modal images and fusion attention mechanism

Electrocardiogram (ECG) signals are of great clinical value for the diagnosis and detection of heart disease. However, current ECG signal recognition research has problems such as large signal morphological differences, unbalanced data distribution, and low recognition accuracy. Therefore, we propose a multi-modal image and fusion attention mechanism based algorithm for automatic ECG signal recognition. The model converts 1D ECG signal data into two-dimensionalization multi-modal image data, retaining the temporal characteristics of the ECG signal while expanding its spatial characteristics. In response to the problems of insufficient spatial feature extraction and redundant temporal feature extraction, we propose a fusion attention module composed of frequency channel attention and convolutional reconstruction unit. Experiments show that the model achieves a classification accuracy of 99.11% and an F1-score of 97.74% on the MIT-BTH dataset. The proposed multi-modal image and fusion attention mechanism improves the algorithm's F1-score by 29%, proving the effectiveness and usability of the algorithm.