Heartbeat Classification Using Convolution Neural Network and Wavelet Transform to Extract Features

In this work, an arrhythmia with abnormal cardiac electrical activity was effectively identified and classified. We propose a new method to classify the five types commonly found in arrhythmia. In the first step, we use wavelet (dmey) transform and 8-point average filter to remove the baseline drift and high frequency noise in the ECG signal. In the second step, centering on the R-wave, intercepts 100 sampling points in front of the R wave, 150 sampling points after the R wave, and a total of 250 sampling points to represent a heartbeat. In the third step, the one-hot label (size 1×5) outputted by the convolutional neural network is combined with the db6 wavelet coefficient (size 1×61) to obtain a total of 66-dimensional features. In the fourth step, the features are normalized and input into the libsvm classifier. We obtained 99,595 heartbeats (MLII) from the MIT-BIH Arrhythmia database and assessed the model performance using a two-fold cross validation. The experimental results show that the total accuracy rate of the classification results is 98.79%. Our proposed algorithm can provide reference for subsequent research.