Arrhythmia detection model modeling method based on multi-view multi-scale fusion network

The invention discloses an arrhythmia detection model modeling method based on a multi-view multi-scale fusion network, and the method comprises the steps: obtaining three-dimensional features of an original ECG signal through a space attention module, a channel attention module and a multi-scale time attention module, and carrying out the residual connection of the features and the original ECG signal, so as to form two mixed features; the two mixed features are input into two parallel networks for deep feature extraction, and the auxiliary information is input into the parallel networks to be fused with the deep features; and extracting the features fused with the auxiliary information in the two paths for further fusion, inputting the fused features into a new multi-layer perceptron to generate a prediction result, taking the prediction result as network output, and using focus loss during training. According to the method, the abnormal features shown in different views when the arrhythmia occurs are conce