Electrocardiosignal denoising method, system and device based on threshold shrinkage network model

Abstract

This invention discloses a method, system, and device for denoising electrocardiogram (ECG) signals based on a threshold contraction network model. The method inputs the ECG signal into a threshold contraction network model for denoising, obtaining a denoised ECG signal output by the model. The denoising process includes: feature extraction from the ECG signal to obtain waveform features; feature extraction from the waveform features to obtain low-frequency signal features, and threshold learning and slope learning from these low-frequency features to obtain a first threshold and a first slope; constructing a threshold function based on the first threshold and the first slope; obtaining a first denoised signal feature based on the threshold function; performing feature enhancement on the first denoised signal feature to obtain a second denoised signal feature; and performing a residual operation between the second denoised signal feature and the ECG signal to obtain the denoised ECG signal. This invention improves denoising performance and reduces the loss of useful signals.

Description

Technical Field

[0001] This invention relates to the field of electrocardiogram (ECG) signal denoising technology, and in particular to an ECG signal denoising method, system, and device based on a threshold contraction network model. Background Technology

[0002] Wavelet-based methods can separate out much noise, but some residual noise remains. This residual noise is highly complex, with an unknown distribution in the frequency domain, making it difficult for existing methods to effectively filter it out. A major drawback of wavelet transform is its inability to perform signal analysis outside the frequency domain. Consequently, many researchers have turned to deep learning to explore more efficient denoising methods. Most current ECG denoising techniques are based on time-domain signals, and their results indicate room for further improvement, especially under noisy conditions such as BW and MA. While residual dense networks are highly effective at extracting features, they also introduce redundant features. Soft thresholding functions, while denoising different types of noisy ECG signals, filter out useful signals, i.e., they lose signal waveform features. These methods are not very effective at denoising ECG signals. Summary of the Invention

[0003] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a method, system, and device for denoising electrocardiogram (ECG) signals based on a threshold contraction network model, which can improve denoising performance and reduce the loss of useful signals.

[0004] In a first aspect, embodiments of the present invention provide a method for denoising electrocardiogram (ECG) signals based on a threshold contraction network model, the method comprising:

[0005] Acquire electrocardiogram (ECG) signals;

[0006] The electrocardiogram (ECG) signal is input into the threshold contraction network model for denoising, resulting in a denoised ECG signal output by the threshold contraction network model; wherein, the denoising process of the threshold contraction network model includes:

[0007] Feature extraction is performed on the electrocardiogram (ECG) signal to obtain ECG waveform features;

[0008] Feature extraction is performed on the waveform features of the electrocardiogram signal to obtain low-frequency signal features, and threshold learning and slope learning are performed on the low-frequency signal features to obtain a first threshold and a first slope;

[0009] Construct a threshold function based on the first threshold and the first slope;

[0010] Based on the threshold function, the low-frequency signal features are subjected to signal denoising, feature refinement, and residual operations to obtain the first denoised signal features;

[0011] The first denoised signal features are enhanced to obtain the second denoised signal features;

[0012] The second denoised signal feature is subjected to a residual operation with the ECG signal to obtain the denoised ECG signal.

[0013] Compared with the prior art, the first aspect of the present invention has the following beneficial effects:

[0014] This method denoises the ECG signal by inputting it into a threshold contraction network model, obtaining the denoised ECG signal output by the threshold contraction network model. Using the threshold contraction network model for ECG denoising improves denoising performance and reduces the loss of useful signals. The threshold contraction network model denoising includes: feature extraction of the ECG signal to obtain waveform features; feature extraction of the waveform features to obtain low-frequency signal features; threshold learning and slope learning of the low-frequency signal features to obtain a first threshold and a first slope; construction of a threshold function based on the first threshold and the first slope; and denoising based on the threshold function. Low-frequency signal features are denoised, refined, and subjected to residual operations to obtain the first denoised signal features. The ECG signal is then denoised using a threshold function constructed based on a first threshold and a first slope. This process removes redundant information, extracts more useful information, reduces the loss of useful signals, and ensures that each ECG signal has its own threshold set, thereby improving the denoising effect. The first denoised signal features are then enhanced to obtain the second denoised signal features. A residual operation is then performed between the second denoised signal features and the ECG signal to obtain the denoised ECG signal. By enhancing the signal features, a more accurate and predictable ECG signal can be obtained.

[0015] According to some embodiments of the present invention, the step of performing signal denoising, feature refinement, and residual operations on the low-frequency signal features based on the threshold function to obtain the first denoised signal features includes:

[0016] The threshold function is used to denoise the low-frequency signal features to obtain the denoised low-frequency signal features.

[0017] The denoised low-frequency signal features are refined to obtain refined low-frequency signal features;

[0018] The refined low-frequency signal features are subjected to residual operation to obtain the first denoised signal features.

[0019] According to some embodiments of the present invention, the step of performing threshold learning on the low-frequency signal features to obtain a first threshold includes:

[0020] The absolute value of the low-frequency signal features is calculated to obtain the first signal feature;

[0021] Redundant information is removed from the first signal feature to obtain the second signal feature;

[0022] The second signal feature is convolved to obtain the third signal feature;

[0023] The third signal feature is subjected to a Sigmoid operation to obtain the fourth signal feature;

[0024] The first signal feature and the fourth signal feature are multiplied by a matrix to obtain the first threshold.

[0025] According to some embodiments of the present invention, the step of performing slope learning on the low-frequency signal features to obtain a first slope includes:

[0026] The low-frequency signal features are sequentially processed through absolute value operation, global average pooling, convolutional layer, and Sigmoid operation to obtain the first slope.

[0027] According to some embodiments of the present invention, the L1 loss function used in the threshold shrinkage network model includes:

[0028]

[0029] Where N represents the number of electrocardiogram signals input into the threshold contraction network model. This represents the denoised electrocardiogram signal output by the threshold shrinkage network model. This indicates a clean electrocardiogram signal.

[0030] According to some embodiments of the present invention, the threshold function is calculated in the following manner:

[0031]

[0032] Where τ represents the first threshold, λ represents the first slope, x represents the waveform characteristics of the electrocardiogram signal, and y represents the low-frequency signal characteristics after denoising.

[0033] According to some embodiments of the present invention, the step of performing feature enhancement on the first denoised signal features to obtain the second denoised signal features includes:

[0034] The first denoised signal features are fused to obtain fused denoised signal features;

[0035] The fused denoised signal features are subjected to feature learning to obtain the denoised signal features after feature learning;

[0036] Perform residual operation on the ECG signal waveform features and the denoised signal features learned from the features to obtain the denoised signal features after residual operation;

[0037] The denoised signal features after the residual operation are enhanced to obtain the second denoised signal features.

[0038] Secondly, embodiments of the present invention also provide an electrocardiogram (ECG) signal denoising system based on a threshold contraction network model, the ECG signal denoising system based on a threshold contraction network model comprising:

[0039] The data acquisition module is used to acquire electrocardiogram (ECG) signals;

[0040] The model denoising module is used to input the electrocardiogram (ECG) signal into the threshold contraction network model for denoising, and obtain the denoised ECG signal output by the threshold contraction network model. The model denoising module includes a first feature extraction submodule, a second feature extraction submodule, a threshold function construction submodule, a signal denoising submodule, a feature enhancement submodule, and a residual operation submodule, wherein:

[0041] The first feature extraction submodule is used to extract features from the electrocardiogram (ECG) signal to obtain ECG signal waveform features;

[0042] The second feature extraction submodule is used to extract features from the waveform features of the electrocardiogram signal to obtain low-frequency signal features, and to perform threshold learning and slope learning on the low-frequency signal features to obtain a first threshold and a first slope.

[0043] The threshold function construction submodule is used to construct a threshold function based on the first threshold and the first slope;

[0044] The signal denoising submodule is used to perform signal denoising, feature refinement and residual operations on the low-frequency signal features according to the threshold function to obtain the first denoised signal features;

[0045] The feature enhancement submodule is used to enhance the features of the first denoised signal to obtain the features of the second denoised signal.

[0046] The residual operation submodule is used to perform a residual operation on the second denoised signal feature and the ECG signal to obtain the denoised ECG signal.

[0047] Thirdly, embodiments of the present invention also provide an electrocardiogram (ECG) signal denoising device based on a threshold contraction network model, including at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which are executed by the at least one control processor to enable the at least one control processor to perform an ECG signal denoising method based on a threshold contraction network model as described above.

[0048] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions for causing a computer to perform an electrocardiogram signal denoising method based on a threshold contraction network model as described above.

[0049] It is understood that the beneficial effects of the second to fourth aspects compared with the related technologies are the same as the beneficial effects of the first aspect compared with the related technologies. Please refer to the relevant description in the first aspect above, which will not be repeated here. Attached Figure Description

[0050] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0051] Figure 1 This is a flowchart of an electrocardiogram signal denoising method based on a threshold shrinkage network model according to an embodiment of the present invention;

[0052] Figure 2 yes Figure 1 The flowchart in step S300;

[0053] Figure 3 This is a schematic diagram of the structure of a threshold shrinkage network model according to an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram of the structure of an improved threshold shrinkage block according to an embodiment of the present invention;

[0055] Figure 5 This is a schematic diagram of the structure of a dense feature learning network according to an embodiment of the present invention;

[0056] Figure 6 This is a schematic diagram of the structure of a simulated self-attention module according to an embodiment of the present invention;

[0057] Figure 7 This is a schematic diagram of the structure of an adaptive learning threshold module according to an embodiment of the present invention;

[0058] Figure 8 This is a dynamic learning slope module according to an embodiment of the present invention;

[0059] Figure 9 This is a schematic diagram illustrating the experimental results of the method according to an embodiment of the present invention;

[0060] Figure 10 This is a structural diagram of an electrocardiogram signal denoising system based on a threshold shrinkage network model according to an embodiment of the present invention;

[0061] Figure 11 yes Figure 10 The structural diagram of the sub-modules in the model denoising module. Detailed Implementation

[0062] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same 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 invention, and should not be construed as limiting the present invention.

[0063] In the description of this invention, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of technical features indicated, or implicitly indicating the order of the technical features indicated.

[0064] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the drawings and are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0065] In the description of this invention, it should be noted that, unless otherwise explicitly defined, terms such as "setting," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0066] Wavelet-based methods can separate out much noise, but some residual noise remains. This residual noise is highly complex, with an unknown distribution in the frequency domain, making it difficult for existing methods to effectively filter it out. A major drawback of wavelet transform is its inability to perform signal analysis outside the frequency domain. Consequently, many researchers have turned to deep learning to explore more efficient denoising methods. Most current ECG denoising techniques are based on time-domain signals, and their results indicate room for further improvement, especially under noisy conditions such as BW and MA. While residual dense networks are highly effective at extracting features, they also introduce redundant features. Soft thresholding functions, while denoising different types of noisy ECG signals, filter out useful signals, i.e., they lose signal waveform features. These methods are not very effective at denoising ECG signals.

[0067] To address the aforementioned problems, this invention denoises the electrocardiogram (ECG) signal by inputting it into a threshold contraction network model, obtaining a denoised ECG signal output by the threshold contraction network model. Denoising the ECG signal using the threshold contraction network model improves denoising performance and reduces the loss of useful signals. The threshold contraction network model denoising process includes: feature extraction of the ECG signal to obtain waveform features; feature extraction of the waveform features to obtain low-frequency signal features; threshold learning and slope learning of the low-frequency signal features to obtain a first threshold and a first slope; constructing a threshold function based on the first threshold and the first slope; and then... The function performs signal denoising, feature refinement, and residual operations on low-frequency signal features to obtain the first denoised signal features. By denoising the ECG signal using a threshold function constructed based on the first threshold and the first slope, some redundant information can be removed, more useful information can be extracted, and the loss of useful signals can be reduced. This ensures that each ECG signal has its own threshold set, thereby improving the denoising effect of the ECG signal. The first denoised signal features are then enhanced to obtain the second denoised signal features. The second denoised signal features are then subjected to residual operations with the ECG signal to obtain the denoised ECG signal. By enhancing the signal features, an ECG signal that better meets expectations can be obtained.

[0068] Reference Figures 1 to 2 This invention provides a method for denoising electrocardiogram (ECG) signals based on a threshold contraction network model. This method includes, but is not limited to, steps S100 to S200 and steps S210 to S260, wherein:

[0069] Step S100: Acquire electrocardiogram (ECG) signal;

[0070] Step S200: Input the ECG signal into the threshold contraction network model for denoising, and obtain the denoised ECG signal output by the threshold contraction network model; wherein, the threshold contraction network model denoising includes:

[0071] Step S210: Extract features from the electrocardiogram (ECG) signal to obtain ECG waveform features;

[0072] Step S220: Extract features from the ECG signal waveform to obtain low-frequency signal features, and perform threshold learning and slope learning on the low-frequency signal features to obtain a first threshold and a first slope;

[0073] Step S230: Construct a threshold function based on the first threshold and the first slope;

[0074] Step S240: Based on the threshold function, perform signal denoising, feature refinement and residual operations on the low-frequency signal features to obtain the first denoised signal features;

[0075] Step S250: Perform feature enhancement on the first denoised signal features to obtain the second denoised signal features;

[0076] Step S260: Perform residual operation on the second denoised signal feature and the ECG signal to obtain the denoised ECG signal.

[0077] In steps S100 to S200 and S210 to S260 of some embodiments, in order to improve denoising performance and reduce the loss of useful signals, this embodiment acquires an electrocardiogram (ECG) signal, inputs the ECG signal into a threshold shrinkage network model for denoising, and obtains a denoised ECG signal output by the threshold shrinkage network model. In order to remove some redundant information and extract more useful information, this embodiment performs feature extraction on the ECG signal to obtain ECG signal waveform features, performs feature extraction on the ECG signal waveform features to obtain low-frequency signal features, and performs threshold learning and slope learning on the low-frequency signal features to obtain a first threshold and a first slope. Based on the first threshold and the first slope, a threshold function is constructed. Based on the threshold function, signal denoising, feature refinement, and residual operation are performed on the low-frequency signal features to obtain a first denoised signal feature. In order to obtain an ECG signal that is more in line with expectations, this embodiment performs feature enhancement on the first denoised signal feature to obtain a second denoised signal feature, and performs a residual operation on the second denoised signal feature and the ECG signal to obtain a denoised ECG signal.

[0078] In some embodiments, the low-frequency signal features are subjected to signal denoising, feature refinement, and residual operations according to a threshold function to obtain first denoised signal features, including:

[0079] A threshold function is used to denoise the low-frequency signal features to obtain the denoised low-frequency signal features.

[0080] The features of the denoised low-frequency signal are refined to obtain the refined low-frequency signal features;

[0081] The refined low-frequency signal features are subjected to residual operation to obtain the first denoised signal features.

[0082] In this embodiment, by using a threshold function to denoise the low-frequency signal features, the features of interest can be learned better, thereby effectively removing noise. Furthermore, by refining the denoised low-frequency signal features, more detailed signal features can be extracted.

[0083] In some embodiments, threshold learning is performed on low-frequency signal features to obtain a first threshold, including:

[0084] The first signal feature is obtained by performing absolute value calculation on the low-frequency signal features.

[0085] Redundant information is removed from the first signal feature to obtain the second signal feature;

[0086] The second signal feature is convolved to obtain the third signal feature;

[0087] The third signal feature is processed by a Sigmoid operation to obtain the fourth signal feature;

[0088] The first threshold is obtained by performing matrix multiplication on the first signal feature and the fourth signal feature.

[0089] In some embodiments, slope learning is performed on low-frequency signal features to obtain a first slope, including:

[0090] The low-frequency signal features are sequentially processed through absolute value operation, global average pooling, convolutional layer, and Sigmoid operation to obtain the first slope.

[0091] In some embodiments, the L1 loss function used by the threshold shrinkage network model includes:

[0092]

[0093] Where N represents the number of ECG signals in the input threshold shrinkage network model. This represents the denoised electrocardiogram (ECG) signal output by the threshold shrinkage network model. This indicates a clean electrocardiogram signal.

[0094] In some embodiments, the threshold function is calculated as follows:

[0095]

[0096] Where τ represents the first threshold, λ represents the first slope, x represents the waveform characteristics of the electrocardiogram signal, and y represents the low-frequency signal characteristics after denoising.

[0097] In this embodiment, by constructing a threshold function using a first threshold and a first slope, the features of interest can be learned more effectively, and the features of noisy signals can be effectively removed.

[0098] In some embodiments, feature enhancement is performed on the first denoised signal features to obtain the second denoised signal features, including:

[0099] The features of the first denoised signal are fused to obtain the fused denoised signal features;

[0100] Feature learning is performed on the features of the fused denoised signal to obtain the denoised signal features after feature learning;

[0101] The residual operation is performed on the waveform features of the electrocardiogram signal and the denoised signal features after feature learning to obtain the denoised signal features after residual operation;

[0102] The denoised signal features after residual operation are enhanced to obtain the second denoised signal features.

[0103] In this embodiment, by enhancing the features of the first denoised signal, an electrocardiogram signal that is more in line with expectations can be obtained.

[0104] To facilitate understanding by those skilled in the art, a set of preferred embodiments is provided below:

[0105] Denoising an ECG signal with a specified signal-to-noise ratio aims to denoise the signal by considering its time domain. The overall network framework diagram of the Threshold Shrink Network Model (ITSN) in this embodiment is shown below. Figure 3 As shown, the one-dimensional signal read from the dataset is directly fed into the network. The Improved Threshold Shrink Block (ITSB) is used as the denoising block of the threshold shrinking network model. The structure diagram of the Improved Threshold Shrink Block is shown below. Figure 4 As shown, the improved threshold shrinking block enhances denoising of one-dimensional ECG signals by considering the extraction of local (Dense Feature Learning Network (DFLN)) and non-local (Residual Learning in ITSB) features and learning the optimal threshold and slope. This threshold shrinking network model is specifically designed for denoising one-dimensional ECG signals. Finally, the threshold shrinking network model outputs a one-dimensional signal that meets the expected parameters, realizing an end-to-end denoising algorithm. The implementation steps are as follows:

[0106] The proposed threshold shrinking network model is inspired by the RDN network. The RDB module in the RDN network is replaced with an Improved Threshold Shrinking Block (ITSB), which includes an adaptive learning threshold module, a dynamic learning slope module, and a threshold function. A relatively clean ECG signal is artificially denoised and then input into the threshold shrinking network model, while the undenoised, relatively clean ECG signal is used as the label. In this embodiment, a shallow feature extraction network consisting of three convolutional layers is first used to extract features from the denoised ECG signal to obtain the ECG signal waveform features. Then, the ITSB acts as a denoising block to learn useful features from the input ECG signal waveform features and remove useless features. First, a dense feature learning network (DFLN) is used to extract low-frequency signal features. The structure of the DFLN is as follows... Figure 5 As shown, an adaptive learning threshold module is then used to adaptively learn the threshold τ, and a dynamic learning slope module is used to adaptively learn the slope (i.e., the slope λ of the threshold function) to effectively remove noise signals from the low-frequency signal features. This is to better learn the features of interest. Next, a simulated self-attention module (SSAM, whose prototype is a self-attention mechanism, but previously it was applied to images of two dimensions or higher. Here, this embodiment changes it to one dimension for one-dimensional signals, but the idea remains the same) is used to fine-tune the signal information. The structure of the simulated self-attention module is as follows. Figure 6 As shown, to extract more refined signal features, multiple ITSBs are stacked to perform deep feature learning and denoising. Finally, a fusion feature representation module (FFR) is used to enhance the signal features, outputting an ECG signal that better matches expectations. The specific process of denoising using the improved threshold shrink block (ITSB) includes:

[0107] Clean ECG signals are used as the target for denoising. Common noise types include white Gaussian noise (WGN), baseline drift (BW), muscle artifacts (MA), and electrode motion (EM). An improved threshold contraction block incorporates an adaptively learned threshold module, a dynamically learned slope module, and a threshold function beneficial for denoising, to achieve efficient denoising. (See reference...) Figure 7The adaptive learning thresholding module includes absolute value operation, global average pooling (GAP), a 1×1 convolutional layer (Conv), and an activation function (Sigmoid). The module first performs an absolute value operation on the input information to obtain the output vector x. Then, it performs average pooling to remove some redundant signals from the input information. Next, it uses a 1×1 convolution operation, which avoids dimensionality reduction; appropriate cross-channel interaction can significantly reduce model complexity while maintaining performance. Then, it uses a Sigmoid operation to scale the output to the range (0, 1) to obtain the output vector y. Finally, it performs a matrix multiplication of the outputs x and y to obtain the output z. This process generates a set of channel thresholds, which are the values ​​required by the threshold function to suppress noise.

[0108] Reference Figure 8 The dynamic learning slope module is similar to the adaptive learning threshold module, also including absolute value, global average pooling (GAP), 1×1 convolutional layer (Conv), and sigmoid activation function. Although the constructed adaptive learning slope module is similar to the adaptive learning threshold module, applying the adaptive learning threshold module to slope inference ensures that the entire module selects the most suitable slope in different channels.

[0109] The threshold function uses the threshold τ obtained by the adaptive learning threshold module and the slope λ learned by the dynamic learning slope module to suppress noise. The calculation formula for the threshold function is as follows:

[0110]

[0111] Where τ represents the first threshold, λ represents the first slope, x represents the waveform characteristics of the electrocardiogram signal, and y represents the low-frequency signal characteristics after denoising.

[0112] After applying the threshold function, a simulated self-attention mechanism is used to further refine the features and extract more useful feature information.

[0113] The Threshold Shrink Network (ITSN) model employs an L1 loss function and several metrics, including signal-to-noise ratio (SNR), root mean square error (PRD), mean square error (MSE), and root mean square error (RMSE). The formulas for calculating the L1 loss function and these metrics are as follows:

[0114]

[0115]

[0116]

[0117]

[0118]

[0119] Where N represents the number of ECG signals in the input threshold shrinkage network model. This represents the denoised electrocardiogram (ECG) signal output by the threshold shrinkage network model. Let u(n) represent the nth raw ECG signal, and let u(n) represent the clean ECG signal. This represents the nth expected electrocardiogram signal.

[0120] For the ECG signal denoising method based on autoencoder, this embodiment uses a threshold shrinkage network model to improve denoising performance while maintaining a relatively complete waveform shape.

[0121] Table 1

[0122]

[0123] Effect demonstration as follows Figure 9 As shown, in Figure 9 middle( Figure 9 The above are comparison images of the ECG signal before and after denoising with 5dB electrode motion (recorded in Table 1). Original_signal is the original ECG signal, noisy_signal is the noisy ECG signal with added 5dB electrode motion, and denosing_signal is the ECG signal after denoising. As can be seen from the above effect images, the denoising effect of this solution and the original waveform shape are well preserved. ECG-GAN in Table 1 is the existing technology "A New ECG Denoising Framework Using Generative Adversarial Network".

[0124] Reference Figures 10 to 11 This invention also provides a denoising system for electrocardiogram (ECG) signals based on a threshold contraction network model. This ECG signal denoising system includes a data acquisition module 100 and a model denoising module 200, wherein:

[0125] Data acquisition module 100 is used to acquire electrocardiogram signals;

[0126] The model denoising module 200 is used to input the electrocardiogram (ECG) signal into a threshold contraction network model for denoising, and obtain the denoised ECG signal output by the threshold contraction network model. The model denoising module includes a first feature extraction submodule 210, a second feature extraction submodule 220, a threshold function construction submodule 230, a signal denoising submodule 240, a feature enhancement submodule 250, and a residual operation submodule 260, wherein:

[0127] The first feature extraction submodule 210 is used to extract features from the electrocardiogram (ECG) signal to obtain the waveform features of the ECG signal;

[0128] The second feature extraction submodule 220 is used to extract features from the waveform features of the electrocardiogram signal, obtain low-frequency signal features, and perform threshold learning and slope learning on the low-frequency signal features to obtain a first threshold and a first slope.

[0129] The threshold function construction submodule 230 is used to construct a threshold function based on the first threshold and the first slope;

[0130] The signal denoising submodule 240 is used to perform signal denoising, feature refinement and residual operation on the low-frequency signal features according to the threshold function to obtain the first denoised signal features;

[0131] The feature enhancement submodule 250 is used to enhance the features of the first denoised signal to obtain the features of the second denoised signal.

[0132] The residual operation submodule 260 is used to perform residual operation on the second denoised signal feature and the ECG signal to obtain the denoised ECG signal.

[0133] It should be noted that since the ECG signal denoising system based on the threshold shrinkage network model in this embodiment is based on the same inventive concept as the ECG signal denoising method based on the threshold shrinkage network model described above, the corresponding content in the method embodiment is also applicable to this system embodiment, and will not be described in detail here.

[0134] This invention also provides an electrocardiogram signal denoising device based on a threshold contraction network model, comprising: at least one control processor and a memory for communicatively connecting to the at least one control processor.

[0135] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0136] The non-transient software program and instructions required to implement the ECG signal denoising method based on a threshold contraction network model in the above embodiments are stored in memory. When executed by a processor, the ECG signal denoising method based on a threshold contraction network model in the above embodiments is executed, for example, the method described above is executed. Figure 1 Method steps S100 to S200 and Figure 2 The method steps S210 to S260.

[0137] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] This invention also provides a computer-readable storage medium storing computer-executable instructions. These instructions are executed by one or more control processors, causing the processors to perform an ECG signal denoising method based on a threshold contraction network model as described in the above-described method embodiments. For example, the method described above... Figure 1 The functions of steps S100 to S200 in the method and Figure 2 The functions of steps S210 to S260 in the method.

[0139] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0140] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for electrocardiosignal denoising based on a threshold shrinkage network model, characterized in that, The ECG signal denoising method based on the threshold shrinkage network model comprises: acquiring an ECG signal; inputting the ECG signal into the threshold shrinkage network model for denoising to obtain an ECG signal denoised by the threshold shrinkage network model; wherein the denoising by the threshold shrinkage network model comprises: performing feature extraction on the ECG signal to obtain ECG signal waveform features; performing feature extraction on the ECG signal waveform features to obtain low-frequency signal features, and performing threshold learning and slope learning on the low-frequency signal features to obtain a first threshold and a first slope; constructing a threshold function according to the first threshold and the first slope; performing signal denoising, feature refinement and residual operation on the low-frequency signal features according to the threshold function to obtain first denoised signal features, wherein the threshold function is calculated in the following manner: wherein τ represents the first threshold, λ represents the first slope, x represents the ECG signal waveform features, and y represents the denoised low-frequency signal features; performing feature enhancement on the first denoised signal features to obtain second denoised signal features; performing residual operation on the second denoised signal features and the ECG signal to obtain the denoised ECG signal.

2. The ECG signal denoising method based on the threshold shrinkage network model according to claim 1, characterized in that, The signal denoising, feature refinement and residual operation on the low-frequency signal features according to the threshold function to obtain first denoised signal features comprise: performing denoising on the low-frequency signal features by using the threshold function to obtain denoised low-frequency signal features; performing feature refinement on the denoised low-frequency signal features to obtain refined low-frequency signal features; performing residual operation on the refined low-frequency signal features to obtain first denoised signal features.

3. The method for electrocardiosignal denoising based on the threshold shrinkage network model according to claim 1, characterized in that, The threshold learning on the low-frequency signal features to obtain a first threshold comprises: performing absolute value operation on the low-frequency signal features to obtain first signal features; performing redundant information removal on the first signal features to obtain second signal features; performing convolution operation on the second signal features to obtain third signal features; performing Sigmoid operation on the third signal features to obtain fourth signal features; performing matrix multiplication on the first signal features and the fourth signal features to obtain the first threshold.

4. The ECG signal denoising method based on the threshold shrinkage network model according to claim 1, characterized in that, The slope learning on the low-frequency signal features to obtain a first slope comprises: sequentially performing absolute value operation, global average pooling, convolution layer and Sigmoid operation on the low-frequency signal features to obtain the first slope.

5. The method for electrocardiosignal denoising based on the threshold shrinkage network model according to claim 1, characterized in that, The L1 loss function used by the threshold shrinkage network model comprises: wherein N represents the number of electrocardio signals input into the threshold shrinkage network model, denotes the de-noised electrocardio signal output by the threshold shrinkage network model, denotes the clean electrocardio signal.

6. The method for electrocardiosignal denoising based on the threshold shrinkage network model according to claim 1, characterized in that, The feature enhancement on the first denoised signal features to obtain second denoised signal features comprises: performing fusion on the first denoised signal features to obtain fused denoised signal features; performing feature learning on the fused denoised signal features to obtain feature-learned denoised signal features; performing residual operation on the ECG signal waveform features and the feature-learned denoised signal features to obtain residual-operated denoised signal features; performing feature enhancement on the residual-operated denoised signal features to obtain second denoised signal features.

7. A system for electrocardiosignal denoising based on a threshold shrinkage network model, characterized in that, The ECG signal denoising system based on the threshold shrinkage network model comprises: The data acquisition module is configured to acquire an electrocardiosignal. The model denoising module is configured to input the electrocardiosignal into the threshold shrinkage network model to perform denoising and obtain a denoised electrocardiosignal output by the threshold shrinkage network model. The first feature extraction submodule is configured to perform feature extraction on the electrocardiosignal to obtain electrocardiosignal waveform features. The second feature extraction submodule is configured to perform feature extraction on the electrocardiosignal waveform features to obtain low-frequency signal features, and perform threshold learning and gradient learning on the low-frequency signal features to obtain a first threshold and a first gradient. The threshold function construction submodule is configured to construct a threshold function according to the first threshold and the first gradient. The signal denoising submodule is configured to perform signal denoising, feature refinement, and residual operation on the low-frequency signal features according to the threshold function to obtain first denoised signal features. The feature enhancement submodule is configured to perform feature enhancement on the first denoised signal features to obtain second denoised signal features. The residual operation submodule is configured to perform residual operation on the second denoised signal features and the electrocardiosignal to obtain a denoised electrocardiosignal. The computer readable storage medium stores computer executable instructions for causing a computer to execute the threshold shrinkage network model based electrocardiosignal denoising method according to any one of claims 1 to 6.

8. A device for electrocardiosignal denoising based on a threshold shrinkage network model, characterized in that The computer readable storage medium stores computer executable instructions for causing a computer to execute the threshold shrinkage network model based electrocardiosignal denoising method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, ​

Images