ARRHYTHMIA RECOGNITION BASED ON SENet-BiLSTM

Deep learning technology shows the problems of low recognition accuracy and weak generalization ability in the arrhythmia classification model. This paper presents an arrhythmia recognition method based on SENet-BiLSTM. This method avoided complex data preprocessing, used a SE-block channel attention mechanism combined with a 6-layer convolutional neural network as a feature extractor, and adopted a bi-directional long short-term memory network as a feature learner to construct a new deep network model to achieve the classification and recognition of five types of arrhythmias. This paper used the MIT-BIH arrhythmia database to test in the inter-patient paradigm and conducted a comparative analysis with CNN, BiLSTM and CNN-BiLSTM. The results show that the overall classification accuracy of the proposed method is 97.25%, which has a better generalization ability.