Multi-noise self-adaptive electrocardiosignal denoising method and system

Abstract

The invention discloses a multi-noise self-adaptive electrocardiosignal denoising method and system, and belongs to the technical field of electrocardiosignal processing. The method comprises the following steps: preprocessing an original noisy ECG signal; carrying out multi-resolution feature extraction and down-sampling on the preprocessed ECG signals by utilizing a mode of combining a multi-layer one-dimensional convolutional network and DWT (Discrete Wavelet Transform) to obtain multi-resolution features; and performing up-sampling and feature reconstruction on the multi-resolution features by using IDWT and a convolutional layer to obtain a de-noised ECG signal. The method strives to overcome the bottleneck in network structure design, efficient and lightweight deployment of the network is realized by improving the denoising and feature retention strategy, the method is fully adapted to a multi-noise scene, and the universality and robustness of ECG denoising are further enhanced.

Description

Technical Field

[0001] The present application proposes a multi-noise adaptive electrocardiogram signal denoising method and system, belonging to the technical field of electrocardiogram signal processing. Background Technique

[0002] In recent years, wearable electrocardiogram (ECG) monitoring devices have been increasingly widely used in remote medical care and personal health management, providing people with relatively convenient and real-time ECG monitoring means. However, the actually collected ECG signals are often affected by various types of noise, such as baseline wander (BW), electromyogram interference (MA), and electrode motion interference (EM), etc. These noises are superimposed together, greatly interfering with the key morphologies and features of the ECG waveform, not only causing difficulties in identifying disease signs, but also possibly leading to deviations in diagnostic results such as arrhythmia.

[0003] ECG signal denoising is a classic problem in the fields of biomedicine and signal processing, aiming to extract accurate cardiac electrical activity information from the original signal affected by various noises for diagnosing diseases such as arrhythmia and myocardial infarction. With the rapid development of wearable devices and Internet of Things technology, it has become a reality to continuously collect ECG signals for a long time using devices such as smart watches and portable electrocardiographs. However, due to factors such as environmental interference, human body movement, and unstable electrode contact, the collected ECG signals are often mixed with various interferences such as baseline wander, electromyogram noise, and electrode motion noise. In the early ECG denoising research, researchers mainly relied on traditional mathematics and signal processing methods, such as low-pass filtering, high-pass filtering, Kalman filtering, wavelet transform, and empirical mode decomposition (EMD), etc., to construct a mapping from the noise signal to the clean signal. These methods usually have the advantages of fast operation speed, sufficient theoretical basis, and strong interpretability. However, in a complex environment with coexisting multiple noises, it is often difficult to suppress the noise while perfectly retaining the detail information in the ECG signal. In recent years, the denoising method based on deep convolutional neural network (CNN) has significantly improved the effect of ECG signal processing. With the help of a large number of learnable convolutional filters, the deep model can automatically extract signal features from the data, achieve precise separation of various noises, and no longer require manual design of complex mapping formulas, thus greatly accelerating the model training speed and improving the denoising efficiency.

[0004] Currently, the relatively mature ECG denoising methods can be mainly divided into two categories: traditional filtering methods and deep learning-based methods. Traditional methods such as low-pass / high-pass filtering, wavelet transform, and adaptive filtering have mature advantages in theory and implementation. However, when dealing with complex noise environments, they often lead to unsatisfactory denoising effects because it is difficult to take into account signal details simultaneously. The denoising method based on deep convolutional neural network realizes the automatic learning of ECG signal features through end-to-end training and has achieved a low root mean square error (RMSE) and a high signal-to-noise ratio (SNR) on public ECG databases. In addition, some research has proposed embedding discrete wavelet transform (DWT) and inverse discrete wavelet transform (IDWT) into CNN to construct a deep wavelet convolutional neural network (DW-CNN) to improve the information loss problem that may occur during downsampling.

[0005] Disadvantages of the prior art: The biggest problem with ECG denoising methods based on traditional theories is that these methods cannot automatically learn and process the key features in electrocardiogram signals. They often rely on manually setting parameters or prior knowledge, which not only greatly increases the calculation and debugging costs but also makes it difficult to comprehensively and accurately restore signal details when facing multiple mixed noises. In addition, although the method based on convolutional neural network can automatically extract features, the features it extracts are mainly limited to the local receptive field of the filter, making it difficult to capture global information. If the network layer is continuously deepened to make up for this defect, it will inevitably bring a large number of parameters and high training and inference costs. On the other hand, although the pure vision transformation network can capture long-range dependencies and obtain global features through block learning, its extremely large number of parameters and the problems of slow operation and unstable training also seriously restrict its application in real-time ECG monitoring and wearable devices. Summary of the Invention

[0006] To address the above problems, this method proposes a high-fidelity ECG denoising method based on deep learning, which focuses on how to fully retain the effective information of the signal in the presence of multiple noises simultaneously, avoiding the limitations brought by only filtering a single or a few specific noise frequency bands in traditional methods. At the same time, although the deep convolutional neural network (CNN) has strong feature extraction ability, operations such as max pooling are often used in the downsampling stage, which is likely to cause the loss of important details. This method aims to overcome this bottleneck in the network structure design, realize the efficient and lightweight deployment of the network by improving the denoising and feature retention strategies, fully adapt to multi-noise scenarios, and further enhance the generality and robustness of ECG denoising.

[0007] To solve the above technical problems, the technical solution adopted in this application is as follows: In the first aspect, this application provides a multi-noise adaptive electrocardiogram signal denoising method, including: Preprocess the original noisy ECG signal; Use a method combining a multi-layer one-dimensional convolutional network and DWT to perform multi-resolution feature extraction and downsampling on the preprocessed ECG signal to obtain multi-resolution features; Use IDWT and convolutional layers to perform upsampling and feature reconstruction on the multi-resolution features to obtain the denoised ECG signal.

[0008] As a further improvement of the present invention, the preprocessing of the original noisy ECG signal includes: Segment the continuously collected original noisy ECG signal according to a fixed number of sampling points to form input data with a unified size; Perform baseline drift correction and amplitude normalization on each segmented signal to remove the DC bias and obtain a normalized ECG signal segment.

[0009] As a further improvement of the present invention, the method of using a multi-layer one-dimensional convolutional network combined with DWT to perform multi-resolution feature extraction and downsampling on the preprocessed ECG signal to obtain multi-resolution features includes: Input the preprocessed ECG signal into multiple convolutional blocks, each convolutional block consisting of several one-dimensional convolutional layers and activation functions to automatically learn the local features of the signal; Introduce a DWT pooling layer after each convolutional block, perform wavelet decomposition on the convolutional output through a predefined wavelet filter, decompose the signal into low-frequency components and high-frequency components, and perform a downsampling operation.

[0010] As a further improvement of the present invention, the decomposition of the signal into low-frequency components and high-frequency components is based on a low-pass filter and a high-pass filter, and the formulas for the low-pass filter and the high-pass filter are:

[0011] In the formula, f L is the low-pass filter, f H is the high-pass filter.

[0012] As a further improvement of the present invention, the operation of the DWT pooling layer is:

[0013] In the formula, X is the input signal, fL and fH are the coefficients of the low-pass filter and the high-pass filter respectively, ↓2 is the downsampling operation, the factor is 2, X cA is the low-frequency component obtained after downsampling, X cD is the high-frequency component obtained after downsampling.

[0014] As a further improvement of the present invention, the use of IDWT and convolutional layer to upsample and reconstruct multi-resolution features to obtain a denoised ECG signal includes: In the IDWT upsampling layer, for the low-frequency component XcA and the high-frequency component XcD, inverse discrete wavelet transform is used for upsampling to reconstruct the features at the original resolution. The inverse transform process is as follows:

[0015] In the formula, X cA is the low-frequency component, X cD is the high-frequency component, IDWT is the inverse discrete wavelet transform, is the reconstructed ECG signal; After IDWT upsampling, a one-dimensional convolutional layer is connected to perform further non-linear transformation and correction on the upsampled features, realizing the upsampling and reconstruction of the signal.

[0016] As a further improvement of the present invention, it further includes a denoised ECG reconstruction step: The corrected features in each layer of the decoder are multiplexed and stacked and fused by using residual skip connections and 1×1 convolutions. The fusion process is as follows:

[0017] In the formula, is the feature obtained from the skip connection, is the finally output denoised ECG signal.

[0018] In a second aspect, the present application provides a multi-noise adaptive ECG signal denoising system, including: A data preprocessing module for preprocessing the original noisy ECG signal; An encoder module for extracting multi-resolution features and downsampling the preprocessed ECG signal by using a multi-layer one-dimensional convolutional network combined with DWT to obtain multi-resolution features; A decoder module for using IDWT and convolutional layer to upsample and reconstruct multi-resolution features to obtain a denoised ECG signal.

[0019] In a third aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the multi-noise adaptive ECG signal denoising method is implemented.

[0020] Fourthly, the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the multi-noise adaptive electrocardiogram signal denoising method is implemented.

[0021] Fifthly, the present application provides a computer program product, and the computer program product includes computer instructions for instructing a computer to execute the multi-noise adaptive electrocardiogram signal denoising method.

[0022] The beneficial effects of the present application compared with the prior art are as follows: The present application proposes a multi-level ECG denoising method that combines the advantages of convolutional neural networks and vision transform networks to overcome the deficiencies of existing methods in automatic feature learning, global information extraction, and computational efficiency. The architecture learns local details in the ECG signal by introducing a deformable convolution module, and uses a non-local attention mechanism and a vision transform network to process the signal in blocks, capture long-range dependencies, and achieve efficient fusion of pixel-level global features. Compared with pure CNN methods, this solution can not only better restore the key details of the ECG signal, but also avoid the explosion of parameters and computational burden caused by deepening the network; compared with pure vision transform network methods, this solution maintains the global feature learning ability while significantly reducing the number of model parameters, improving the running speed and training stability, and thus is more suitable for clinical real-time monitoring and resource-constrained wearable device applications. Description of the Drawings

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following introduces the drawings of the related technical solutions in the embodiments of the present application or the prior art. It should be understood that the drawings introduced below are only for conveniently and clearly presenting some embodiments of the technical solutions in the present application, and those skilled in the art can also obtain other drawings based on these drawings without creative efforts.

[0024] Figure 1 It is a flowchart of a multi-noise adaptive electrocardiogram signal denoising method provided by the present application; Figure 2 It is a schematic diagram of an ECG denoising system based on a deep wavelet convolutional neural network. Detailed Embodiments

[0025] Embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application and should not be construed as a limitation of the present application. For the step numbers in the following embodiments, they are only set for the convenience of elaboration and explanation, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0026] In the description of the present application, unless otherwise clearly defined, words such as "set", "installed", "connected", etc. should be understood in a broad sense. Those skilled in the art can reasonably determine the specific meanings of the above words in the present application in combination with the specific content of the technical solution.

[0027] The present application proposes a multi-noise adaptive electrocardiogram (ECG) signal denoising method, which is an improved ECG denoising method based on a deep wavelet convolutional neural network (DW-CNN). The method specifically includes the following steps: S1. Preprocess the original noisy ECG signal; specifically including: S11. Segment the continuously collected original noisy ECG signal according to a fixed number of sampling points to form input data with a unified size. S12. Perform baseline drift correction and amplitude normalization on each segmented signal to remove the DC bias and obtain a normalized ECG signal segment.

[0028] S2. Use a method combining a multi-layer one-dimensional convolutional network and DWT to perform multi-resolution feature extraction and downsampling on the preprocessed ECG signal to obtain multi-resolution features; specifically including: S21. Input the preprocessed ECG signal into multiple convolutional blocks. Each convolutional block consists of several one-dimensional convolutional layers and activation functions to automatically learn the local features of the signal. S22. Introduce a DWT pooling layer after each convolutional block. Perform wavelet decomposition on the convolutional output through a predefined wavelet filter to decompose the signal into low-frequency components and high-frequency components, and perform a downsampling operation.

[0029] In the above solution, decomposing the signal into low-frequency components and high-frequency components is based on a low-pass filter and a high-pass filter. The formulas for the low-pass filter and the high-pass filter are:

[0030] In the formula, f L is the low-pass filter, f H is the high-pass filter.

[0031] Among them, the DWT pooling layer operation is as follows:

[0032] In the formula, X is the input signal, fL and fH are the low-pass filter and high-pass filter coefficients respectively, ↓2 is the downsampling operation with a factor of 2, X cA is the low-frequency component obtained after downsampling, and X cD is the high-frequency component obtained after downsampling.

[0033] S3. Use IDWT and the convolutional layer to upsample and reconstruct the multi-resolution features to obtain the denoised ECG signal. Specifically, it includes: S31. In the IDWT upsampling layer, for the low-frequency component XcA and the high-frequency component XcD, use the inverse discrete wavelet transform to upsample and reconstruct the features at the original resolution. The inverse transform process is as follows:

[0034] In the formula, X cA is the low-frequency component, X cD is the high-frequency component, IDWT is the inverse discrete wavelet transform, is the reconstructed ECG signal; S32. After IDWT upsampling, connect to a one-dimensional convolutional layer to perform further non-linear transformation and correction on the upsampled features, and realize the upsampling and reconstruction of the signal.

[0035] On the basis of the above solution, preferably, the present invention further includes: S4, denoised ECG reconstruction step: Adopt the method of residual skip connection and 1×1 convolution to multiplex and stack and fuse the corrected features in each layer of the decoder. The fusion process is as follows:

[0036] In the formula, is the feature obtained from the skip connection, is the finally output denoised ECG signal.

[0037] This method is based on the encoder-decoder framework, introduces a multi-level residual skip connection and 1×1 convolution feature fusion module, and realizes the end-to-end denoising and reconstruction of the ECG signal. The dynamic correction mechanism of the residual skip connection: Through the cross-layer connection between each layer of the encoder and the corresponding layer of the decoder, multi-scale features are transmitted and adaptively corrected. The lightweight design of cross-layer feature fusion: Use 1×1 convolution to compress the channel dimension, reduce the computational complexity, and at the same time retain the key feature information.

[0038] Based on the above solution, the residual skip connection of the present invention transmits low-level detailed features (such as QRS wave peaks) and suppresses the amplification of high-frequency noise. The 1×1 convolution reduces the number of parameters for cross-layer fusion and is suitable for the deployment of embedded ECG devices. The frequency-domain loss function avoids over-smoothing and retains key features for clinical diagnosis.

[0039] This method adopts an improved DWT pooling and IDWT upsampling strategy, and uses the Haar wavelet to perform real-time decomposition and reconstruction of one-dimensional ECG features, thus effectively avoiding the problem of key information loss that may be caused by traditional max pooling operations. The overall system is designed based on an encoder-decoder framework. The encoder consists of several convolutional layers and DWT pooling layers connected in series to extract and compress features in the ECG signal layer by layer. The decoder correspondingly uses convolutional layers and IDWT upsampling layers to gradually restore the feature resolution and finally output the denoised ECG signal. In addition, this solution has multi-noise adaptability, and the network automatically learns noise features through end-to-end training without distinguishing different noise types such as BW, MA, and EM.

[0040] This application also provides a multi-noise adaptive electrocardiogram signal denoising system. The ECG denoising system based on DW-CNN is as Figure 1 shown, and mainly includes a data preprocessing module, an encoder module, and a decoder module. This method first performs segmentation and normalization processing on the original noisy ECG signal through the data preprocessing module, and then inputs the preprocessed signal into the encoder module. In the encoder module, a multi-resolution feature extraction and downsampling of the ECG signal are performed by combining a multi-layer one-dimensional convolutional network and DWT. Then, in the decoder module, upsampling and feature reconstruction are achieved by using IDWT and convolutional layers to obtain a high-quality denoised ECG signal.

[0041] The following is a detailed description of the specific implementation solutions for each module. Figure 2 It is a schematic diagram of an ECG denoising system based on a deep wavelet convolutional neural network.

[0042] Figure 2 In this invention, first, the collected noisy electrocardiogram signal is preprocessed, then multi-resolution features are extracted by using multi-layer convolution and discrete wavelet pooling in the encoder stage, and finally, the signal is gradually reconstructed in the decoder stage through inverse discrete wavelet upsampling and convolutional layers, and a high-fidelity denoised electrocardiogram signal is obtained through residual fusion. The specific description is as follows: The data preprocessing module is mainly used to preprocess the collected original noisy ECG signals to ensure the consistency and high quality of the subsequent network input data. In the specific implementation process, the continuously collected ECG signals are first segmented according to a fixed number of sampling points to form input data with a unified size. Subsequently, baseline wander (BW) correction and amplitude normalization are performed on each segmented signal to remove the DC bias and ensure that the original features of the signal are not distorted. After the above processing, the standardized ECG signal segments output by the preprocessing module will be used as the input of the subsequent network module.

[0043] The encoder module includes an input layer, a convolutional layer, and a pooling layer based on discrete wavelet transform; the encoder module uses a method of alternately connecting a multi-layer one-dimensional convolutional network and a discrete wavelet transform (DWT) pooling layer to extract features and downsample the input ECG signal to achieve multi-resolution feature representation. Specifically, the encoder module first inputs the preprocessed ECG signal into multiple convolutional blocks, each of which consists of several one-dimensional convolutional layers (Conv1D) and activation functions to automatically learn the local features of the signal. Then, a DWT pooling layer is introduced after each convolutional block, and the convolutional output is wavelet decomposed through a predefined wavelet filter to decompose the signal into a low-frequency component (cA) and a high-frequency component (cD), while implementing the downsampling operation. Its low-pass filter and high-pass filter are respectively defined as:

[0044] In the formula, f L is the low-pass filter, f H is the high-pass filter.

[0045] For the input feature X, the DWT pooling operation can be expressed as:

[0046] In the formula, X is the input signal, fL and fH are the low-pass filter and high-pass filter coefficients respectively, ↓2 is the downsampling operation, the factor is 2, X cA is the low-frequency component obtained after downsampling, X cD is the high-frequency component obtained after downsampling.

[0047] Through this multi-level encoder structure, the network can extract rich multi-resolution features in the ECG signal, providing sufficient information for subsequent signal reconstruction.

[0048] The decoder module is used to reconstruct the low-resolution features output by the encoder into a high-resolution ECG signal. Its main implementation scheme includes two parts: an upsampling layer based on inverse discrete wavelet transform and a convolutional layer. The specific processing methods include: First, in the IDWT upsampling layer, for the low-frequency component XcA and high-frequency component XcD obtained in the encoder, inverse discrete wavelet transform (IDWT) is used for upsampling to reconstruct the features at the original resolution. The inverse transform process can be expressed as:

[0049] where, X cA is the low-frequency component, X cD is the high-frequency component, IDWT is the inverse discrete wavelet transform, is the reconstructed ECG signal; This operation is the inverse process of DWT and can completely restore the signal information after downsampling. Subsequently, a one-dimensional convolutional layer is connected after IDWT upsampling to perform further non-linear transformation and correction on the upsampled features to reduce the reconstruction error, thereby realizing signal upsampling and reconstruction layer by layer.

[0050] As a further improvement, on this basis, in order to fully reuse and fuse the features of each layer, this application further sets a denoising ECG reconstruction function in the decoder module. Specifically, this reconstruction module uses residual skip connections and 1×1 convolutions to reuse and stack-fuse the corrected features in each layer of the decoder. The fusion process can be expressed as:

[0051] where, is the feature obtained from the skip connection, is the finally output denoised ECG signal.

[0052] The ECG signal output by this module is the high-resolution reconstructed signal after efficient denoising processing.

[0053] This application's system proposes to use discrete wavelet transform (DWT) to replace the traditional max-pooling operation, which can better retain the details of the ECG signal during the downsampling process. At the same time, this application uses inverse discrete wavelet transform (IDWT) to replace the traditional transposed convolution or upsampling method, thereby achieving high-precision restoration of waveform details.

[0054] Furthermore, this application has an end-to-end encoder-decoder structure, which can adaptively process various types of noise without pre-classifying the input data or performing additional feature engineering. At the same time, since the entire segment of the ECG signal is directly processed, it avoids the instability of QRS detection in a strong-noise environment, thereby reducing the cumulative error and automatically learning the deep denoising mapping relationship.

[0055] Compared with the existing optimal technologies, the present application adopts an end-to-end encoder-decoder structure, does not require normalization in preprocessing, can directly process the entire segment of ECG signals, and avoids the cumulative problems caused by QRS detection errors. By introducing DWT pooling and IDWT upsampling techniques to replace traditional pooling and deconvolution operations, signal details are effectively retained, while the number of parameters and computational complexity are significantly reduced, achieving near real-time denoising.

[0056] Therefore, the present invention was tested on the publicly available database MIT-BIH, and trained and tested by adding different types and intensities of noise to clean ECG signals. Under different signal-to-noise ratio conditions such as 0 dB, 1.25 dB, and 5 dB, this method has an average SNR improvement of about 0.3 - 1.2 dB compared with traditional filtering methods or ordinary CNNs, and has better ECG signal denoising performance.

[0057] As a further improvement, the Haar wavelet used in the encoder or decoder can be replaced with Daubechies or Coiflet wavelets, and the convolution kernel size and number of network layers can be adjusted according to actual hardware conditions; or by introducing an attention mechanism, adopting a bottleneck structure or grouped convolution, and using knowledge distillation and hierarchical training strategies, the denoising ability of a large model can be transferred to a lightweight network.

[0058] The third objective of the embodiments of the present application is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above multi-noise adaptive electrocardiogram signal denoising method is implemented. It also includes a communication interface and a bus.

[0059] The fourth objective of the embodiments of the present application is to provide a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above multi-noise adaptive electrocardiogram signal denoising method is implemented.

[0060] The fifth objective of the embodiments of the present application is to provide a computer program product, which includes computer instructions that direct a computer to execute the above multi-noise adaptive electrocardiogram signal denoising method.

[0061] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0062] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or steps for implementing the functions specified in multiple blocks of one block or multiple blocks.

[0063] This application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0064] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks of one block or multiple blocks.

[0065] Obviously, the described embodiments are only partial embodiments of this application, rather than all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of this application or make equivalent replacements. Any modification or equivalent replacement without departing from the spirit and scope of this application shall be covered by the protection scope of the claims of this application.

Claims

1. A multi-noise adaptive electrocardiogram signal denoising method, characterized in that: include: Preprocessing the original noisy ECG signal; By combining a multi-layer one-dimensional convolutional network with DWT, multi-resolution feature extraction and downsampling are performed on the preprocessed ECG signal to obtain multi-resolution features; IDWT and convolutional layers are used to upsample and reconstruct multi-resolution features to obtain denoised ECG signals.

2. A multi-noise adaptive ECG signal denoising method according to claim 1, characterized in that: The preprocessing of the original noisy ECG signal comprises: The continuously acquired original noisy ECG signal is segmented according to a fixed number of sampling points to form input data of uniform size; Each segmented signal is subjected to baseline drift correction and amplitude normalization to remove the DC offset and obtain a standardized ECG signal segment.

3. The multi-noise adaptive ECG signal denoising method according to claim 1, characterized in that: The method of combining a multi-layer one-dimensional convolutional network with a DWT is used to extract and downsample the preprocessed ECG signal to obtain multi-resolution features, including: The preprocessed ECG signal is input into multiple convolution blocks, each of which consists of several one-dimensional convolution layers and activation functions to automatically learn the local features of the signal; A DWT pooling layer is introduced after each convolution block, and the convolution output is subjected to wavelet decomposition through a predefined wavelet filter to decompose the signal into low-frequency components and high-frequency components, and then a downsampling operation is performed.

4. The multi-noise adaptive ECG signal denoising method according to claim 3, characterized in that: The decomposition of the signal into low-frequency components and high-frequency components is based on low-pass filters and high-pass filters. The formulas of low-pass filters and high-pass filters are: In the formula, f L is a low-pass filter, f H is a high pass filter.

5. The multi-noise adaptive ECG signal denoising method according to claim 3, characterized in that: The DWT pooling layer operation is: Where X is the input signal, fL and fH are the coefficients of the low-pass filter and high-pass filter respectively, ↓2 is the downsampling operation with a factor of 2, and X cA is the low-frequency component obtained after downsampling, X cD is the high frequency component obtained after downsampling.

6. The multi-noise adaptive ECG signal denoising method according to claim 1, characterized in that: The method uses IDWT and convolutional layers to upsample and reconstruct multi-resolution features to obtain denoised ECG signals, including: In the IDWT upsampling layer, the low-frequency component XcA and the high-frequency component XcD are upsampled using the inverse discrete wavelet transform to reconstruct the features of the original resolution. The inverse transformation process is: In the formula, X cA is the low frequency component, X cD High frequency components, IDWT is inverse discrete wavelet transform, is the reconstructed ECG signal; After IDWT upsampling, a one-dimensional convolution layer is connected to perform further nonlinear transformation and correction on the upsampled features to achieve signal upsampling and reconstruction.

7. The multi-noise adaptive ECG signal denoising method according to claim 1, characterized in that: A denoised ECG reconstruction step is also included: The corrected features in each layer of the decoder are reused and stacked by using residual skip connection and 1×1 convolution. The fusion process is as follows: In the formula, is the feature obtained from the skip connection, is the final output denoised ECG signal.

8. A multi-noise adaptive ECG signal denoising system, characterized in that: include: A data preprocessing module, used for preprocessing the original noisy ECG signal; An encoder module is used to extract and downsample the preprocessed ECG signal using a multi-resolution feature extraction and downsampling method in combination with a multi-layer one-dimensional convolutional network and a DWT to obtain a multi-resolution feature; The decoder module is used to upsample and reconstruct multi-resolution features using IDWT and convolutional layers to obtain denoised ECG signals.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the multi-noise adaptive electrocardiogram signal denoising method as claimed in claim 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the multi-noise adaptive electrocardiogram signal denoising method according to claim 8 is implemented.

Images