Research On ECG Signal Denoising Algorithm Based On Multi-layer Noise Reduction Autoencoder

Heart diseases seriously threaten human physical and mental health,especially the highest prevalence among middle-aged and elderly people over 50 years old.Therefore,the prevention and diagnosis of cardiovascular diseases have become a top priority.Analysing whether the heart is diseased or not is currently a more effective means of preventing and treating heart diseases.However,the ECG signal is an electrophysiological signal with low signal-to-noise ratio and susceptible to interference,so the denoising of the ECG signal has become a research hotspot in the field of modern medicine.However,the traditional ECG signal denoising algorithm is good at removing linear and smooth ECG signal noise,and the denoising effect is not ideal when facing nonlinear and non-stationary ECG signal noise.In order to solve this problem,this paper studies a denoising algorithm of ECG signal based on noise reduction autoencoder.The main contents include:(1)The noise reduction autoencoder can extract the input signal features and has the characteristics of robustness to noise(DAE).The noise reduction autoencoder is used to construct a shallow neural network to reduce the noise of the ECG signal.(2)ECG signal denoising based on multi-layer noise reduction autoencoder(SDAE).In order to solve the problem that the shallow network model cannot extract the deep signal characteristics of the ECG signal,multiple noise reduction autoencoders are used to construct a deep learning network to complete the ECG signal noise reduction task.In the network training,unsupervised training is used to complete the network pre-training to obtain the network initialization parameters,and the overall network parameters of the network are optimized according to the error back propagation algorithm.According to different network structures and different training samples,the optimal network structure and appropriate training samples are selected to improve the denoising effect of the deep neural network model on ECG signals.(3)ECG signal denoising based on the multi-layer contraction noise reduction automatic encoder(SCDAE).In order to solve the problem that the input data interferes with the neural network when there are too many training samples,and affects the denoising effect of the neural network model,the shrinking autoencoder has the characteristics of hidden layer disturbance invariance and the noise reduction autoencoder has the ability to resist noise.Features to improve the overall robustness of the noise reduction network.The experimental results show that: the average signal-to-noise ratio obtained by thetraditional ECG signal denoising algorithm is 11 d B;the average signal-to-noise ratio obtained by the noise reduction automatic encoder(DAE)algorithm is 13 d B;the multilayer noise reduction automatic encoder(SDAE)algorithm The average signal-to-noise ratio obtained is 18 d B;the average signal-to-noise ratio obtained by the SCDAE algorithm is21 d B.Compared with the DAE algorithm and the traditional ECG signal denoising algorithm,the deep learning algorithm has a better noise reduction effect on the nonlinear and non-stationary ECG signal noise.Moreover,it is found through experiments that the ECG signal obtained after the algorithm processing in this paper basically retains the characteristic form of the original signal,and has strong anti-interference ability to noise.