An ECG signal classification method based on the combination of CNN and GRU based on self-encoding mode

Abstract

An ECG signal classification method based on the combination of CNN and GRU in the self-encoding mode, by extracting the most representative features in the original signal, using CNN+GRU for feature extraction, saving space and saving a lot of training space, among which The GRU (Gated Recurrent Unit) used on the one hand solves the problem of gradient disappearance and gradient explosion due to RNN training. On the other hand, it has one less gate than LSTM, which is easier to calculate and can improve training efficiency. The advantage of GRU is that when When there are few training samples, it can be used to prevent overfitting. When there are many training samples, it can also save a lot of training time, and can improve the learning efficiency of the network and the accuracy of ECG signal recognition.

Description

technical field [0001] The invention relates to the technical field of ECG signal classification, in particular to a method for classifying ECG signals based on the combination of CNN and GRU in an autoencoder mode. Background technique [0002] Electrocardiographic (ECG) signals are a widely used noninvasive method of detection of underlying cardiac conditions. The ECG signal is the most basic indicator for doctors to evaluate the patient's heart condition, but because the physiological signal is affected by the internal changes of the individual, for example, the electrode position and noise will affect the waveform of the signal, and even the ECG of healthy subjects The signal, in different situations, the shape of QRS complex, P wave and R-R interval will not be the same between different beats, and the ECG signal of the same type of arrhythmia is very different between different stages in the same patient. There may be obvious changes, and the difference in ECG signals...

AI Technical Summary

Problems solved by technology

These limit the applicability of the methods, which may encounter large variations when classifying ECG signals from new subjects, making them often vary widely in accuracy and efficiency