| Electrocardiogram(ECG)can reflect the changes in human heart potential.It is one of the most widely used techniques in the clinical diagnosis of cardiovascular disease.The first lead ECG is getting more and more attention because of its great convenience and scientificity.For the ECG classification,algorithm flow often includes four steps: signal preprocessing,heartbeat segmentation,feature extraction and classification recognition.The existing algorithms always distinguish the heartbeat classification problem from the whole ECG signal classification problem.Besides,the accuracy of ECG classification also needs to be improved.For the first lead ECG heartbeat classification problems,this paper proposes a heartbeat classification algorithm based on deep convolution neural network(CNN).One-dimensional convolutional neural network is designed for the single-lead ECG signals.The final network structure has the characteristics of multiple layers,multi-scale convolution kernels and small parameter size.The classification algorithm is fast and can achieve real-time performance.The heartbeat classification experiment is carried out on the INCART database.The experiment results show that the proposed algorithm can classify the first lead ECG heartbeats very well,and prove the effectiveness and efficiency of the algorithm.The performance of the algorithm can be further applied to wearable devices and remote monitoring area.For the first lead whole ECG signal classification problems,this paper designs an ECG signal classification convolutional neural networks fused with heartbeat feature,called HF-CNN(Heartbeat Fusion CNN).This network is based on the whole ECG signal convolutional neural network(CNN1)and the heartbeat convolutional neural network(CNN2).The network is constructed by merging the heartbeat features of CNN2 middle layer into the classification network of CNN1.Experiment results on the CCDD database show that the heartbeat features can effectively enhance the ECG signal characteristics.Compared with other algorithms,the performance of this algorithm has been improved significantly,and significant classification results have been reached. |
