Two-stage ECG signal noise reduction method based on convolutional autoencoder

Abstract

The invention discloses a two-stage electrocardiographic signal noise reduction method based on a convolutional self-encoder. A two-stage electrocardiographic signal noise reduction method based on a convolutional autoencoder, comprising: the data sampling rate of the electrocardiographic signal prepared is 360HZ, the input length of the electrocardiographic signal is fixed to a uniform length, and different degrees are added to the electrocardiographic signal. The corresponding label output is prepared for each set of data, and the output is a clean ECG signal; the ECG signal noise reduction stage; the ECG signal detail reconstruction stage. Beneficial effects of the present invention: the method does not need to convert the ECG signal into the time-frequency domain or other expression functions, and can directly learn the noise reduction process of the ECG signal without generating the Gibbs effect; the method does not need to obtain Noise reference signal; this method can better restore the detail part and characteristic signal of the original signal through the process of signal noise reduction and detail reconstruction.

Description

technical field [0001] The invention relates to the field of electrocardiographic signals, in particular to a two-stage electrocardiographic signal noise reduction method based on a convolutional autoencoder. Background technique [0002] Electrodes are placed in different parts of the human body and connected to the positive and negative poles of the ECG machine through lead wires. This circuit connection method for recording ECG is called ECG lead. An electrocardiogram is essentially a time-voltage graph of the potential changes during the heartbeat. In a normal cardiac cycle, a typical ECG waveform is composed of a P wave, a QRS complex, a T wave, and a U wave that may be seen in 50% to 75% of ECGs [1]. The P wave corresponds to atrial depolarization, the QRS complex corresponds to ventricular depolarization, and the T wave corresponds to the process of ventricular repolarization. like figure 1 shown (refer to the national standard YY 0782-2010 / IEC60601-2-51:2003). At...

AI Technical Summary

Problems solved by technology

[0005] 2. The frequency of baseline drift is generally between 0.05-2Hz, which is caused by breathing, electrified motor or subject movement;

[0007] 4. Power frequency interference, caused by defects in sampling equipment or public power grid radiation

[0022] 1. The ECG signal after noise reduction is prone to Gibbs effect, which affects the shape of Q wave and S wave

[0023] 2. Manual intervention is required to select the threshold and threshold function, resulting in a decline in generalization ability

[0032] The existing ECG signal noise reduction algorithm needs to obtain noise reference signals, which are difficult to obtain with the ECG signal acquisition system;

[0033] The existing ECG signal denoising algorithm cannot restore the details of the high-quality original signal well

Images