Magnetocardiogram signal noise reduction method based on global feature extraction and peak law learning

Abstract

The invention discloses a magnetocardiogram signal noise reduction method based on global feature extraction and peak law learning. A coding-decoding framework is adopted, the feature extraction capacity of the model is enhanced through jump residual connection, a peak attention mechanism is designed, and the position of an R peak in a magnetocardiogram signal is focused to capture a key mode. Besides, through global self-attention network modeling, the long-term dependence and global regularity of the MCG signal are learned, so that the real signal is accurately restored under the condition of low signal-to-noise ratio. Different from most methods which only use a single heartbeat signal, the method utilizes an MCG sequence with a longer time span to further capture the global rule and subtle change of the heartbeat. Experimental results show that the method is remarkably superior to an existing method in the aspects of noise reduction precision, robustness and signal recovery capability, and effective technical support is provided for low-cost and high-precision cardiovascular disease diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of magnetocardiogram signal denoising, and particularly to a magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning. Background Art

[0002] Electrocardiogram (ECG) signals are one of the most commonly used tools for diagnosing cardiovascular diseases. Doctors can evaluate the electrical activity of the heart and identify potential abnormalities, such as arrhythmias and myocardial infarctions, by analyzing ECG signals. However, ECG signals have certain limitations. For example, they are affected by the quality of electrode contact and have low spatial resolution, making it difficult to detect weak electrophysiological abnormalities in the early stages of some heart diseases.

[0003] In contrast, magnetocardiogram (MCG), as a non-contact imaging technology, can significantly reduce the noise interference caused by tissue conduction, and at the same time has low signal distortion and higher spatial resolution. This gives MCG an obvious advantage in the accuracy of cardiac source localization. Especially in the early diagnosis of cardiovascular diseases, MCG can capture subtle electrophysiological changes that cannot be detected by ECG, such as early minute changes in ischemic myocardium, thus showing higher sensitivity. However, despite the significant advantages of MCG technology, its high equipment cost limits its wide clinical application. To solve this problem, the present invention uses a relatively low-cost tunneling magnetoresistance (TMR) sensor to collect MCG signals. Although the TMR sensor reduces the cost, the collected signals are often affected by complex noises, including baseline drift, power frequency noise, and Gaussian white noise. The noise problem of MCG signals is similar to that of ECG. The complex noise components make it difficult for simple denoising methods to effectively remove the noise, and may even cause the loss of key waveform features, thus affecting subsequent analysis and applications.

[0004] In recent years, deep learning technology has made remarkable progress in the field of signal processing, and deep learning-based denoising methods have gradually become a research hotspot. Such methods can capture the complex relationship between signals and noise by learning features in large-scale data. Compared with traditional denoising methods, deep learning has stronger adaptability, can automatically learn noise features and suppress them, thus achieving more accurate denoising effects. For example, Prateek Singh et al. [Singh P, Sharma A. Attention-based convolutional denoising autoencoder for two-lead ECG denoising and arrhythmia classification[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1-10.] proposed an attention-based convolutional denoising autoencoder model, which uses skip connections and attention modules to reconstruct ECG signals under high-noise conditions. However, this method mainly focuses on local signal features and fails to fully consider the relevance of global signals. In response to this, Zhu et al. [Zhu Y, Zhu D, Liu J. RA-LENet: R-Wave Attention and Local Enhancement for Noise Reduction in ECG Signals[C] / / 2024 International Joint Conference on Neural Networks (IJCNN). IEEE, 2024: 1-9.] proposed an ECG signal denoising method that combines the R-wave attention mechanism and local enhancement, which captures key information from a global perspective and significantly improves the signal restoration effect.

[0005] Although existing deep learning methods have made great progress in ECG signal denoising, they still face many challenges when dealing with more complex MCG signals, especially in effectively learning global patterns and multi-peak signal laws in long time series. Traditional methods usually focus on feature extraction of local signals while ignoring global time dependence and the regularity of multi-peak signals, which limits the further improvement of denoising effects.

[0006] In view of the above problems, the present invention proposes a method for denoising MCG signals by combining global feature extraction and peak pattern learning. This method can capture the global correlation and regularity of signals in long time series, and is particularly suitable for complex structures containing multiple peak signals. Different from most methods that only take a single heartbeat signal as input, the present invention uses a longer time-span MCG sequence to fully capture the overall regularity of heartbeats. In view of the prominent characteristics of the R peak in MCG signals, a peak attention mechanism is introduced to focus on key signals, further improving the denoising effect. In addition, the MCG signals are modeled by a global self-attention network to learn their long-term dependencies and pattern regularities. Especially in a low signal-to-noise ratio environment, it can more accurately remove noise and restore the true waveform of the signals. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for denoising magnetocardiogram signals based on global feature extraction and peak pattern learning in view of the deficiencies of the prior art.

[0008] To achieve the above purpose, the present invention provides a method for denoising magnetocardiogram signals based on global feature extraction and peak pattern learning, including the following steps: S1: Collect magnetocardiogram signals through a tunneling magnetoresistive sensor in a magnetic shielding room as target signals; at the same time, collect magnetocardiogram signals affected by environmental noise through a tunneling magnetoresistive sensor in an unshielded room as noisy signals; construct a noisy data set containing target signals and noisy signals; S2: Construct a denoising model including an input layer, a globally enhanced encoding / decoding module, and an output layer; both the input layer and the output layer are composed of a fully connected layer and a convolutional layer for dimensional conversion of signal features; the globally enhanced encoding / decoding module is composed of multiple encoding / decoding blocks for capturing and enhancing global features in signals, where each encoding / decoding block contains two global feature extraction modules; the global feature extraction module is used to extract global features in signals, project the feature information of the encoding block into the corresponding decoding block through residual skip connections, and capture key peak information in signals using the peak attention mechanism; S3: Input the noisy signals collected in step S1 into the denoising model constructed in step S2 to start the training process of the model; use the noisy data set to continuously optimize the parameters of the model by adjusting the mean square error of the loss function and using the optimizer function, so that the output signal of the model gradually approaches the target signal, thereby achieving denoising; S4: Input the magnetocardiogram signals to be tested into the trained model to obtain the denoised signals.

[0009] Furthermore, random noise addition and random deletion operations are performed on the noisy signals to simulate noise interference and signal loss or sensor failure.

[0010] Further, in step S2, the input layer includes a fully connected layer and a convolutional layer to convert the dimension of the signal features from both global and local perspectives.

[0011] Further, in step S2, the globally enhanced encoding / decoding module consists of four encoding / decoding blocks, and each encoding / decoding block contains two global feature extraction modules; the global feature extraction module is mainly composed of a signal position encoder, a normalization layer, a peak attention mechanism, and a multi-layer perceptron.

[0012] Further, in the encoding stage, the input signal is first converted into a feature representation containing position information through the signal position encoder for feature extraction operations to introduce sequence position information; in the decoding stage, the output signal of the encoder first passes through the global signal generated by the global feature extraction module, and then is input into the decoding module for restoration operations.

[0013] Further, in step S2, the output layer includes a fully connected layer and a convolutional layer to convert the dimension of the signal features from both global and local perspectives.

[0014] Further, in the noise reduction model, the noisy signal is dimensionally transformed through the input layer, and then the transformed signal is input into the globally enhanced encoding / decoding module. Among them, there are four blocks for encoding and decoding respectively, and each encoding / decoding block contains two global feature extraction modules; during the encoding / decoding process, the feature information obtained by encoding is projected into the corresponding decoding block respectively using residual skip connections to avoid feature loss during the decoding process; in the global feature extraction module, the captured global peak information is incorporated into each encoding / decoding block using the peak attention mechanism to enhance the model's learning of global key information, thereby achieving noise reduction. Finally, the feature dimension of the signal is restored through the output layer to obtain the denoised signal.

[0015] To achieve the above object, the present invention also provides a magnetocardiogram signal noise reduction device based on global feature extraction and peak pattern learning, including one or more processors for implementing the above magnetocardiogram signal noise reduction method based on global feature extraction and peak pattern learning.

[0016] To achieve the above object, the present invention also provides an electronic device including a memory and a processor, and the memory is coupled to the processor; wherein, the memory is used to store program data, and the processor is used to execute the program data to implement the above magnetocardiogram signal noise reduction method based on global feature extraction and peak pattern learning.

[0017] To achieve the above object, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning is implemented.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Traditional methods are often limited to the extraction of local signal features and are difficult to effectively capture the global correlation in long time series. This method designs a feature extraction mechanism from a global perspective, models the long-term dependence relationship of the signal through a global self-attention network, so that key waveform features can be efficiently restored in a noisy environment. In addition, different from most models that only process single heartbeat signals, this method is based on a 10-second long time series input, captures the global regularity of heartbeats as a whole, and provides a new solution idea for denoising complex magnetocardiogram signals. Through experiments and evaluations, the method of the present invention has achieved remarkable effects in practical applications.

[0019] (2) Regarding the significance of the R peak in the magnetocardiogram signal, the present invention innovatively proposes a peak attention mechanism, effectively focuses on multiple peak signals in the long time series, and significantly improves the accuracy and reliability of denoising. This design not only strengthens the retention of key waveform features but also enhances the robustness of the model under low signal-to-noise ratio conditions.

[0020] (3) Compared with the benchmark model, the denoising ability of the present invention shows obvious improvement in terms of accuracy, robustness, and reliability, can better handle complex anomaly detection tasks, and has broad application prospects. In practical applications, the model can be further optimized and adjusted according to specific requirements to achieve the best denoising effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the architecture diagram of the method of the present invention; Figure 2 is the flowchart of the method of the present invention; Figure 3 is the schematic diagram of the target signal and the signal collected by the sensor in the embodiment of the present invention; Figure 4 is the schematic diagram of the signal with randomly deleted noise and the signal after denoising and restoration of the signal with randomly deleted noise in the embodiment of the present invention; Figure 5 is the schematic diagram of the signal with randomly added noise and the signal after denoising and restoration of the signal with randomly added noise by the model in the embodiment of the present invention; Figure 6 is the structural schematic diagram of the device of the present invention; Figure 7 is the schematic diagram of an electronic device of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] Next, in combination with the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] See Figure 1 and Figure 2 , the present invention provides a magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning; including the following steps: S1: Data preprocessing of magnetocardiogram signals: First, use a portable or non-magnetically shielded magnetocardiogram measurement device (TMR sensor) to collect magnetocardiogram signals in a magnetically shielded room. Since there is no external environmental noise in this environment and the collected signals have filtered out the DC component, these signals are considered pure target signals for model evaluation.

[0024] At the same time, in order to obtain noisy signals, the present invention uses the same sensing device to collect magnetocardiogram signals again on the same individual in a non-shielded room. Since there is electromagnetic interference in these environments (such as computers, wireless devices, medical instruments, vehicles, etc.), the collected signals naturally contain different degrees of environmental noise. These noisy signals maintain individual consistency with the target signals but are affected by the real environment, thus constituting a noisy dataset for model training.

[0025] In addition, in order to verify the effectiveness of the present invention, two different noise interference simulations are performed on the noisy signals captured by the sensor: random noise addition and random deletion. Specifically, by randomly adding noise to the collected noisy signals, strong Gaussian white noise is added between three and seven seconds of the signal to simulate noise interference, while the random deletion operation randomly selects and sets that part of the signal to 0 within three to seven seconds to simulate the situation of signal loss or sensor failure. Through these noise interference simulations, the robustness and denoising effect of the present invention in the face of different types of noise interference can be evaluated.

[0026] S2: The model is mainly composed of an input layer, a globally enhanced encoding / decoding module, a global feature extraction module, and an output layer stacked together. The globally enhanced encoding / decoding module is used to capture and enhance the global features in the signal. The feature extraction module enhances the model's understanding of key signals by learning the global patterns in the signal, especially key information such as R peaks. First, the noisy magnetocardiogram signal is dimensionally transformed through the input layer, and then the transformed magnetocardiogram signal is input into the globally enhanced encoding / decoding module. Among them, there are four blocks for both encoding and decoding, and each encoding / decoding block contains two global feature extraction modules respectively. During the encoding / decoding process, the feature information obtained by encoding is projected into the corresponding decoding block using residual skip connections respectively, so as to avoid feature loss during the decoding process. In addition, in the global feature extraction module, the peak attention mechanism is mainly used to integrate the captured global peak information into each encoding / decoding block, enhancing the model's learning of global key information, thus effectively realizing the noise reduction function. Finally, the feature dimension of the signal is restored through the output layer to obtain the denoised signal.

[0027] First, define the input magnetocardiogram signal and , where N represents the number of samples, d represents the dimension of each sample, and T represents the transpose operation. Secondly, in the proposed model, the input layer and the output layer are mainly composed of a fully connected layer and a convolutional layer, which are mainly used to transform the dimension of the signal features from both global and local perspectives; the globally enhanced encoding / decoding module is composed of two globally feature extraction modules in series. Among them, the global feature extraction module is mainly composed of a signal position encoder, a normalization layer, a peak attention mechanism, and a multi-layer perceptron. It mainly compresses and then restores the transformed signal, and at the same time uses the peak attention mechanism to focus on the key information in the signal, thereby reconstructing and generating a new signal to achieve the purpose of denoising. The specific operations are as follows: (2.1) Input layer: In order to effectively transform high-dimensional data, the input layer contains a fully connected layer and a convolutional layer, so as to transform the dimension of the signal features from both global and local perspectives, thereby improving the computational efficiency of the entire network. First, take the signal S as the input, and perform feature dimension transformation through the input layer to obtain the output , represents the feature dimension after transformation by the input layer.

[0028] (2.2)Global enhanced encoding / decoding module: The present invention designs an encoding / decoding module, in which there are four encoding modules and four decoding modules respectively, and each module contains two global feature extraction modules. The global feature extraction module is mainly composed of a signal position encoder, a normalization layer, a peak attention mechanism, and a multi-layer perceptron. In addition, during the encoding / decoding process, the feature information obtained by encoding is projected into the corresponding decoding block respectively by using residual skip connections, so as to avoid feature loss caused during the decoding process.

[0029] First, in the encoding stage, the input signal S is first converted into a feature representation containing position information through the signal position encoder , for further feature extraction operations to introduce sequence position information. The formula is as follows:

[0030] where PE represents the position encoding module, which converts the signal S into a feature representation embedding position information . Then, the global feature extraction module extracts features from a global perspective through the normalization layer, the multi-layer perceptron, and the peak attention mechanism. Specifically, the global feature extraction is expressed as:

[0031] where MLP represents the multi-layer perceptron, Norm is the normalization operation, GFE represents the global feature extraction module, and , represents the signal feature representation learned after the peak attention mechanism, the normalization operation, and the residual connection. PAM represents the peak attention mechanism. In this process, the peak attention mechanism (PAM) is used to identify the features with important position information in the input signal to enhance the model's ability to capture global dependencies. The calculation formula of the peak attention mechanism is:

[0032] where Q, K, and V are different linear projections of the input signal respectively, and PLmask is the position mask, which is specially designed to encode the position information of the peaks in the signal, and can dynamically enhance the attention weights related to these key positions, so as to effectively capture the significant features in the signal.

[0033] First, the output signal obtained by encoding through the four encoder modules is defined as E. In the decoding stage, the output signal of the encoder first passes through the global signal generated by the global feature extraction module , and then it is input into the decoding module for restoration operation:

[0034] where, , represents the signal feature representation learned after the peak attention mechanism, normalization operation, and residual connection. The enhancement mechanism mainly passes the signal obtained by encoding into the corresponding decoding module through the way of residual skip connection. is the signal feature captured by the corresponding encoding module. After passing through four decoding modules, the output is obtained.

[0035] (2.3) Output layer: To effectively restore the signal, the output layer contains a fully connected layer and a convolutional layer, which convert the dimensions of the signal features from both global and local perspectives to improve the computational efficiency of the entire network. The signal features learned by the decoding module are input into the output layer, and the final output signal is obtained.

[0036] S3: Input the noisy signal used for training in step S1 into the noise reduction model in step S2, and through the last output layer of the model, the denoised magnetocardiogram signal is obtained. During the entire model training process, by adjusting the loss function Mean Squared Error (MSE), optimizer function, and learnable hyperparameters, finally find the combination of hyperparameters that makes the model perform best on the noisy dataset, continuously optimize the parameters of the model, and make the output signal of the model gradually approach the target signal, so as to achieve denoising.

[0037]

[0038] Among them, represents the signal output by the model, represents the target signal, and N represents the length of the signal (i.e., the number of samples).

[0039] S4: Input the magnetocardiogram signal to be tested into the trained model to obtain the denoised signal.

[0040] Adopt a variety of widely used metric indicators SNR, RMSE to evaluate the performance of the model. The established model outperforms the existing models in terms of the SNR and RMSE indicators of the dataset compared with several other models.

[0041] The flowchart is as Figure 2As shown. After obtaining the signal output by the model through the above operations, widely used metric indicators, namely, Root Mean Squared Error (RMSE) and Signal-to-Noise Ratio (SNR), were adopted to evaluate the performance of the model. Among them, RMES can measure the error between the predicted value and the true value of the model, and SNR can measure the ratio of the useful information of the signal to the noise. The larger the value, the higher the signal quality and the smaller the noise interference. The following are the calculation formulas for the evaluation index formulas:

[0042]

[0043] Among them, represents the signal output by the model, represents the target signal, and N represents the length of the signal (i.e., the number of samples).

[0044] See Figure 3 、 Figure 4 and Figure 5 wherein this Figure 3 shows the target signal and the signal collected by the sensor, Figure 4 shows the signal with 3 - 7 seconds of noise randomly deleted and the signal after denoising and restoring the signal with 3 - 7 seconds of noise randomly deleted, Figure 5 shows the signal with 3 - 7 seconds of noise randomly added and the signal after denoising and restoring the signal with 3 - 7 seconds of noise randomly added by the model. It can be seen from Figures 3 - 5 that the model of the present invention has a certain effectiveness in denoising and restoring.

[0045] Corresponding to the foregoing embodiment of the magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning, the present invention also provides an embodiment of a magnetocardiogram signal denoising device based on global feature extraction and peak pattern learning.

[0046] See Figure 6 The magnetocardiogram signal denoising device based on global feature extraction and peak pattern learning provided by the embodiment of the present invention includes one or more processors for implementing the magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning in the above embodiment.

[0047] Embodiments of the magnetocardiogram signal denoising device based on global feature extraction and peak pattern learning of the present invention can be applied to any device with data processing capabilities, and such a device with data processing capabilities can be a device or apparatus such as a computer. The device embodiments can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. At the hardware level, as Figure 6 shown, it is a hardware structure diagram of any device with data processing capabilities where the magnetocardiogram signal denoising device based on global feature extraction and peak pattern learning of the present invention is located. In addition to Figure 6 the processor, memory, network interface, and non-volatile memory shown, usually according to the actual functions of any device with data processing capabilities where the device in the embodiment is located, other hardware may also be included, which will not be elaborated here.

[0048] For the implementation processes of the functions and roles of each unit in the above device, please refer to the implementation processes of the corresponding steps in the above method for details, which will not be elaborated here.

[0049] For the device embodiments, since they basically correspond to the method embodiments, relevant parts can refer to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0050] Corresponding to the foregoing embodiments of the magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning, embodiments of the present application further provide an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning as described above. As Figure 7 shown, it is a hardware structure diagram of any device with data processing capabilities where the magnetocardiogram signal denoising method based on global feature extraction and peak pattern learning provided by the embodiments of the present application is located. In addition to Figure 7In addition to the processor, memory, DMA controller, disk, and non-volatile memory shown, any device with data processing capabilities where the device in the embodiment is located may generally include other hardware according to the actual functions of the device with data processing capabilities, which will not be elaborated here.

[0051] Corresponding to the embodiments of the magnetocardiogram signal noise reduction method based on global feature extraction and peak pattern learning described above, an embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the magnetocardiogram signal noise reduction method based on global feature extraction and peak pattern learning in the above embodiments is implemented.

[0052] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or will be output.

[0053] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0054] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed by the present invention are within the scope of protection of the present invention.

Claims

1. A method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning, characterized in that: The following steps are involved: S1: The magnetocardiogram signal is collected in a magnetically shielded room by a tunnel magnetoresistive sensor as the target signal; at the same time, the magnetocardiogram signal interfered by environmental noise is collected in a non-shielded room by a tunnel magnetoresistive sensor as the noisy signal; a noisy data set containing the target signal and the noisy signal is constructed; S2: construct a denoising model including an input layer, a globally enhanced encoding / decoding module and an output layer; the input layer and the output layer are both composed of a fully connected layer and a convolutional layer for dimensional transformation of signal features; the globally enhanced encoding / decoding module is composed of a plurality of encoding / decoding blocks for capturing and enhancing global features in the signal, wherein each encoding / decoding block contains two global feature extraction modules; The global feature extraction module is used to extract global features in the signal, project the feature information of the encoding block to the corresponding decoding block through residual skip connection, and use the peak attention mechanism to capture the key peak information in the signal; S3: Input the noisy signal collected in step S1 into the denoising model constructed in step S2 to start the model training process; use the noisy data set to continuously optimize the model parameters by adjusting the mean square error of the loss function and using the optimizer function so that the output signal of the model gradually approaches the target signal, thereby achieving denoising; S4: Input the magnetic cardiotonic signal to be tested into the trained model to obtain a denoised signal.

2. The method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning according to claim 1, characterized in that: The noisy signal is subjected to random noise addition and random deletion operations to simulate noise interference and signal loss or sensor failure.

3. The method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning according to claim 1, characterized in that: In step S2, the input layer includes a fully connected layer and a convolutional layer to transform the dimensions of signal features from a global and local perspective.

4. The method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning according to claim 1, characterized in that: In step S2, the globally enhanced encoding / decoding module consists of four encoding / decoding blocks, each of which contains two global feature extraction modules; the global feature extraction module is mainly composed of a signal position encoder, a normalization layer, a peak attention mechanism and a multi-layer perceptron.

5. The method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning according to claim 4, characterized in that: In the encoding stage, the input signal is first converted into a feature representation containing position information through a signal position encoder for feature extraction operations to introduce sequence position information; in the decoding stage, the output signal of the encoder is first converted into a global signal generated by a global feature extraction module, and then input into the decoding module for restoration operations.

6. The method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning according to claim 1, characterized in that: In step S2, the output layer includes a fully connected layer and a convolutional layer to transform the dimensions of the signal features from a global and local perspective.

7. The method for denoising magnetic cardiogram signals based on global feature extraction and peak rule learning according to claim 1, characterized in that: In the denoising model, the noisy signal is transformed in dimension through the input layer, and then the transformed signal is input into a globally enhanced encoding / decoding module, wherein encoding and decoding have four blocks respectively, and each encoding / decoding block contains two global feature extraction modules; in the encoding / decoding process, the feature information obtained by encoding is projected into the corresponding decoding blocks respectively by using residual jump connections, thereby avoiding feature loss caused during the decoding process; in the global feature extraction module, the peak attention mechanism is used to integrate the captured global peak information into each encoding / decoding block, and the model's learning of global key information is enhanced, thereby achieving denoising, and finally the feature dimension of the signal is restored through the output layer to obtain the denoised signal.

8. A device for reducing the noise of magnetic cardiogram signals based on global feature extraction and peak rule learning, characterized in that: It comprises one or more processors for implementing the method for reducing the noise of magnetic cardio signals based on global feature extraction and peak regularity learning as described in any one of claims 1 to 7.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the magnetic cardiosignal denoising method based on global feature extraction and peak rule learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for reducing the noise of magnetic cardiogram signals based on global feature extraction and peak rule learning as described in any one of claims 1 to 7 is implemented.

Images