Electrocardiogram classification method based on improved SSA and DMSANet

Abstract

The invention relates to an electrocardiogram classification method based on improved SSA and DMSANet, and the method comprises the steps: obtaining an electrocardiogram signal, and carrying out the noise reduction processing through an improved SSA algorithm; a DMSANet is constructed, the DMSANet comprises a backbone network, sub-processing modules which are arranged in parallel and have different scales and a multi-scale feature fusion module, and the electrocardiogram signals subjected to noise reduction processing are classified through the DMSANet. According to the method, the noise reduction effect is remarkably improved, and the accuracy and the signal quality of the ECG signal are effectively improved; a backbone network based on a ResNet architecture in a DMSANet learns and extracts features from ECG signals, a multi-subnet and multi-scale feature fusion strategy is beneficial to capturing and integrating feature information of different scales, and the classification performance and robustness of the ECG signals are improved; by combining an improved SSA algorithm and a DMSANet architecture, the overall reliability and the application effect of classification research based on ECG signals are improved, and the method is particularly suitable for the field of biomedical signal processing.

Description

Technical Field

[0001] The present invention relates to the technical field of general image data processing or generation, and in particular to an electrocardiogram classification method based on improved SSA and DMSANet in the field of signal processing and deep learning. Background Art

[0002] Cardiovascular disease (CVD) is a major challenge faced by people worldwide. It includes a variety of heart and blood vessel related diseases, such as coronary heart disease, cerebrovascular disease, rheumatic heart disease and other cardiovascular diseases. According to statistics, about 17.9 million people die from cardiovascular disease each year, making it one of the leading causes of death from disease worldwide.

[0003] Early diagnosis and intervention are crucial to reducing cardiovascular disease-related mortality. In terms of diagnosis, the electrocardiogram (ECG) is a major diagnostic tool that records the electrical signals generated by the heart at each beat. These signals can reflect the rhythm and functional status of the heart. Doctors can analyze ECG to detect abnormal heart rhythms, myocardial ischemia, signs of myocardial infarction and other pathological changes, thereby helping to diagnose and treat cardiovascular diseases.

[0004] However, in the process of collecting ECG signals, it is often interfered by various noises, including but not limited to motion artifacts, baseline drift, electromyogram (EMG) interference, static interference, power line interference (50Hz) and breathing. These interference factors seriously affect the clarity and accuracy of ECG signals, thereby reducing the practical value of ECG in clinical diagnosis; in addition, analyzing ECG records is usually time-consuming and prone to human errors, which may lead to misdiagnosis and can only achieve minimal accuracy.

[0005] Therefore, it is urgent to develop an ECG filtering and classification algorithm that can effectively suppress various noise interferences, ensure signal accuracy and integrity, and realize effective signal classification processing. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides an electrocardiogram classification method based on improved SSA and DMSANet, which effectively handles the noise and classification problems faced by ECG.

[0007] The technical solution adopted by the present invention is an electrocardiogram classification method based on improved SSA and DMSANet. The method obtains an electrocardiogram signal and performs noise reduction processing on the electrocardiogram signal using an improved SSA algorithm; constructs DMSANet, including a backbone network, sub-processing modules of different scales arranged in parallel, and a multi-scale feature fusion module, and uses DMSANet to obtain classification of the electrocardiogram signal after noise reduction.

[0008] Preferably, the improved SSA algorithm includes a signal decomposition method, a principal component confirmation method and a signal reconstruction method which are arranged in sequence. After the electrocardiogram signal is decomposed by the signal decomposition method, the principal component confirmation method is used to obtain the effective part of the electrocardiogram signal, and the electrocardiogram signal is reconstructed by the signal reconstruction method.

[0009] Preferably, the signal decomposition method comprises the following steps:

[0010] S1.1.1 Let the ECG signal be a sequence of length N, and use embedding operation to map it into a vector sequence of length L, and map it into the L×K space to obtain the trajectory matrix X, 1 <L<N;

[0011] S1.1.2 performs singular value decomposition on the trajectory matrix X.

[0012] Preferably, the principal element confirmation method comprises the following steps:

[0013] S1.2.1 construct a new index, and reconstruct the matrix of the electrocardiogram signal based on the new index;

[0014] S1.2.2 mapping the reconstructed matrix into a set of reconstructed one-dimensional time series by diagonal averaging method, and defining each one-dimensional time series as a principal element of the original electrocardiogram signal;

[0015] S1.2.3 Determine and select effective principal components;

[0016] S1.2.4 Extract features based on effective principal components.

[0017] Preferably, S1.2.3 comprises the following steps:

[0018] S1.2.3.1 Take any pivot as sample data, and whether it is selected as the target pivot is used as the sample label, and the sample label is 0 or 1;

[0019] S1.2.3.2 Apply the logistic regression model to the sample data, combine the Sigmod function with the linear regression function, and calculate the probability that any sample data belongs to the sample label;

[0020] S1.2.3.3 Use the gradient descent algorithm and the cross-entropy loss function to extract key features from each single principal component as the input of the logistic regression model, and determine whether the current principal component is a valid principal component based on the output of the logistic regression model.

[0021] Preferably, the dimensions of the features include frequency domain, nonlinearity, and time domain. The features here are features extracted from the principal component, including but not limited to signal mean, root mean square value, signal kurtosis value, etc., which can be determined as needed.

[0022] Preferably, the backbone network includes a CNN backbone network, a convolutional layer and two maximum pooling layers arranged in sequence, and three groups of combinations of first convolutional layers and CBAM modules and four groups of combinations of second convolutional layers and CBAM modules are arranged in sequence between the two maximum pooling layers.

[0023] Preferably, the DMSANet includes 2 sub-processing modules; any sub-processing module includes 3 groups of third convolutional layers and CBAM modules arranged in sequence, 2 groups of fourth convolutional layers and CBAM modules, and a global pooling layer.

[0024] Preferably, the loss optimization is performed based on the focal loss.

[0025] The present invention provides an electrocardiogram classification method based on improved SSA and DMSANet, which obtains electrocardiogram signals and performs noise reduction processing on the electrocardiogram signals using an improved SSA algorithm; constructs DMSANet, including a backbone network, sub-processing modules of different scales arranged in parallel, and a multi-scale feature fusion module, and classifies the noise-reduced electrocardiogram signals using DMSANet.

[0026] The beneficial effects of the present invention are:

[0027] (1) When processing the complete ECG signal, the improved SSA algorithm is used for noise reduction. The logistic regression algorithm is introduced in the principal component recombination stage, and the principal component recombination method is improved to automatic recombination, which significantly improves the noise reduction effect and effectively improves the accuracy and signal quality of the ECG signal.

[0028] (2) The backbone network based on the ResNet architecture in DMSANet learns and extracts features from ECG signals. The multi-subnet and multi-scale feature fusion strategies help capture and integrate feature information at different scales, thereby improving the classification performance and robustness of ECG signals.

[0029] (3) Combining the improved SSA algorithm and DMSANet architecture, the overall reliability and application effect of classification research based on ECG signals are improved, which is particularly suitable for the field of biomedical signal processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the method of the present invention;

[0031] Figure 2 It is an operation flow chart from the initial electrocardiogram signal to the noise-reduced electrocardiogram signal in the present invention;

[0032] Figure 3 It is a schematic block diagram of the structure of DMSANet in the present invention;

[0033] Figure 4 For the corresponding Figure 3 The network structure parameter diagram;

[0034] Figure 5 It is a structural schematic diagram of the CBAM module in the present invention;

[0035] Figure 6 is the F1 score of the method of the present invention. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto.

[0037] like Figure 1 As shown, the present invention relates to an electrocardiogram classification method based on improved SSA and DMSANet, wherein the method obtains an electrocardiogram signal, performs noise reduction processing on the electrocardiogram signal using an improved SSA algorithm; constructs DMSANet, including a backbone network, sub-processing modules of different scales arranged in parallel, and a multi-scale feature fusion module, and classifies the noise-reduced electrocardiogram signal using DMSANet.

[0038] The method is described in detail below.

[0039] like Figure 2 As shown, the improved SSA algorithm includes a signal decomposition method, a principal component confirmation method and a signal reconstruction method which are arranged in sequence. After the electrocardiogram signal is decomposed by the signal decomposition method, the principal component confirmation method is used to obtain the effective part of the electrocardiogram signal, and the electrocardiogram signal is reconstructed by the signal reconstruction method.

[0040] In the present invention, an improved SSA algorithm is used to reduce the noise of ECG signals. The initial SSA algorithm consists of signal decomposition and signal reconstruction.

[0041] In the signal decomposition process, the initial ECG signal is decomposed after embedding operation and singular value decomposition, realizing the mapping of the signal in the independent dimension in the high-dimensional space, so that the effective components and noise components in the signal are separated;

[0042] Then, the single principal component diagonal averaging, feature extraction and logistic regression algorithms are used to select the corresponding effective components in the high-dimensional space and eliminate the noise components.

[0043] In the signal reconstruction stage, the target signal is reconstructed, and the obtained ECG signal is noise-reduced.

[0044] Specifically, the signal decomposition method includes the following steps:

[0045] S1.1.1 Let the electrocardiogram signal be a sequence of length N, and map it into a sequence of vectors of length L by embedding operation, and map it into the space of L×K to obtain the trajectory matrix X, where 1 < L < N;

[0046] Here, take x = (x1,..., x N ) as the research object, where N > 2, and x is a non-zero sequence, that is, there is at least one integer i between 1 and N such that x i ≠0;

[0047] Let the integer L (1 < L < N) be the window length, and K = N - L + 1. The embedding operation maps the original time series (electrocardiogram signal series) into a sequence of vectors of length L and maps it into the space of L×K to obtain the trajectory matrix X, satisfying

[0048]

[0049] S1.1.2 Perform singular value decomposition on the trajectory matrix X;

[0050] In the present invention, the trajectory matrix is subjected to singular value decomposition and expressed as the sum of first-order biorthogonal matrices;

[0051] Let the covariance matrix S = XX T , λ1, λ2... λ L be the eigenvalues of S and λ1 ≥... ≥ λ L ≥0, and the singular value decomposition of X satisfies

[0052]

[0053] where d = rank(X), and U and V are the left and right singular matrices of the trajectory matrix X respectively;

[0054] Here, λ1, λ2... λ L are the eigenvalues of the covariance matrix, specifically the values of the middle diagonal matrix in the following decomposition.

[0055] In fact, the signal reconstruction stage includes two steps: grouping and diagonal averaging. In the grouping stage, the principal element sequence of the index set {1, 2,..., d} can be divided into several discontinuous subsets according to actual needs, and the purpose of grouping is to find the components suitable for electrocardiogram signal reconstruction, which is realized by the principal element confirmation method here;

[0056] Specifically, the main element confirmation method includes the following steps:

[0057] S1.2.1 Construct a new index, and reconstruct the matrix of the electrocardiogram signal based on the new index, that is, select the corresponding effective components for recombination in the high-dimensional space, eliminate the noise components, so as to reconstruct the target signal and achieve the purpose of noise reduction;

[0058] Let I = {i1,…i p}(p < d) be the corresponding new index, then the matrix corresponding to the ECG can be expressed as

[0059]

[0060] S1.2.2 Map the reconstructed matrix X group to a group of reconstructed one-dimensional time series x recon , expressed as

[0061]

[0062] where A8 = {(l,k): l + k = s + 1, 1 ≤ l ≤ L, 1 ≤ k ≤ K}, l and k correspond to L and K in X group , representing the variables of the index, and N(A s ) is the number of elements in the set A s ;

[0063] After reconstructing the diagonal average, the original signal is given. Taking x as the filtered electrocardiogram signal, the original one-dimensional time series signal x can be expressed as:

[0064] x = x recon(1) + x recon(2) + … + x recon(d)

[0065] where x recon(n) is reconstructed from the grouped matrix X group(n) (n = 1,…,d), and x recon(n) is defined as the main element of the original electrocardiogram signal; the original signal is restored by the superposition of multiple main elements, and each x recon(n) here is one of the important components of the signal.

[0066] S1.2.3 Judge and screen the effective main elements;

[0067] The selection of principal component sequence is crucial in the signal reconstruction stage of singular spectrum analysis. However, the selection of principal component sequence usually depends on the experience of researchers and lacks a universal standard. In addition, for different types of signals, the position of the principal component of the effective component in the sequence is not fixed, which makes it difficult to perform effective noise reduction only by the principal component with a fixed position, and also limits the automated execution of the singular spectrum analysis algorithm. On this basis, a logistic regression algorithm is proposed. By constructing feature engineering based on ECG signals and noise signals, the principal components obtained after signal decomposition are judged to determine which principal components are the effective components of the ECG signals, and the automated selection of principal components is realized, thereby reducing manual intervention and improving the automation level of the singular spectrum analysis algorithm.

[0068] Specifically, S1.2.3 includes the following steps:

[0069] S1.2.3.1 Take any pivot x recon(i) As sample data, corresponding to whether it is selected as the target principal element as the sample label, the sample label is 0 or 1, y i ∈{0,1}; each pair of sample data and sample label forms a sample category pair (x recon(i) ,y i );

[0070] S1.2.3.2 Apply the logistic regression model to the sample data, and transform the Sigmod function g(z) = 1 / (1+ze) and the linear regression function f(x) = ω t Combine x+b to calculate the probability that any sample data belongs to the sample label.

[0071]

[0072] Among them, the parameter weight ω and intercept b are estimated using the maximum likelihood method, and the cross entropy loss function can be obtained after deformation.

[0073]

[0074] S1.2.3.3 Use the gradient descent algorithm based on the cross entropy loss function. Specifically, based on the cross entropy loss function, by minimizing the difference between the prediction and the actual value and minimizing the loss function, the optimal solution of the weight ω and the intercept b is obtained, and the key features are extracted from each single principal component as the input of the logistic regression model to improve the classification performance of the model;

[0075] Taking into account the complexity of the model, features are extracted from a single principal component signal according to the frequency domain, nonlinearity, time domain and other dimensions. In the principal component selection step of the singular spectrum analysis, whether the current principal component is a valid principal component is determined based on the output of the logistic regression model to achieve automatic principal component selection.

[0076] S1.2.4 Extract features based on effective principal components and extract a certain number of characteristics from the effective principal component signal, including but not limited to the signal mean, signal kurtosis value, frequency of the main energy spectrum, sample entropy of the signal, etc.

[0077] like Figure 3 , Figure 4 As shown in the figure, in the specific implementation process, the deep multi-scale attention network (DMSANet) is used to perform ECG classification operations. First, the backbone network is used for learning and feature extraction to obtain the original feature representation. The features are sent to two sub-processing modules respectively to perform feature extraction at different scales. Each sub-processing module focuses on processing feature maps of a specific scale to capture more detailed local features and broader global features. Subsequently, after feature extraction at different scales, these feature maps are fused together.

[0078] The backbone network includes a CNN backbone network, a convolutional layer and two maximum pooling layers arranged in sequence, and three groups of first convolutional layers and CBAM module combinations and four groups of second convolutional layers and CBAM module combinations are arranged in sequence between the two maximum pooling layers.

[0079] In the present invention, in a specific implementation process, the number of channels of the convolution layer and the first convolution layer is 64, and the number of channels of the second convolution layer is 128, so as to capture more features.

[0080] Specifically, ResNet has excellent performance and is used to build the backbone network; Figure 5 As shown in the figure, CBAM is an attention mechanism module for enhancing the performance of convolutional neural network (CNN). It improves the perception ability of the model by introducing channel attention and spatial attention in CNN, thereby improving the performance without increasing the complexity of the network.

[0081] The LeakyReLU activation function solves the "dead neuron" problem of the traditional ReLU function, provides a small gradient in the negative interval, maintains the gradient flow of the network, and improves the stability of the training process.

[0082] The DMSANet includes two sub-processing modules; any sub-processing module includes three groups of third convolutional layers and CBAM modules, two groups of fourth convolutional layers and CBAM modules, and a global pooling layer.

[0083] In this step, the feature map of the model is passed to two sub-processing modules, which perform feature extraction at different scales. Each sub-network focuses on processing feature maps of a specific scale to obtain more refined local features and more macroscopic global features. By extracting and processing features at different scales, the model can more comprehensively understand the multi-level information in the data, thereby improving the ability to recognize complex patterns and thus improving the accuracy of classification or detection.

[0084] In the present invention, in a specific implementation process, the number of channels of the third convolutional layer is 256, and the number of channels of the fourth convolutional layer is 512, which are used to capture features of more different scales.

[0085] After feature extraction, the feature maps obtained from different sub-processing modules are fused to integrate features at each scale for more effective loss optimization.

[0086] The present invention uses focal loss for loss optimization to emphasize difficult-to-classify samples and reduce the impact of easy-to-classify samples on loss calculation.

[0087] Focal Loss(p tj )=-α t (1-p tj ) γ logp tj

[0088] Among them, p tj is the predicted probability of the model, α t It is a balancing factor used to balance the weights of positive and negative classes, and γ is an adjustment factor, usually a positive number, used to adjust the weights of samples that are difficult to classify.

[0089] In ECG datasets, class imbalance is common, and common diseases are usually much more numerous than uncommon diseases. Therefore, the network needs to focus not only on the most common diseases, but also on less common diseases. Focal Loss is used to solve the class imbalance problem, reduce the weight of samples that are easy to classify, and increase the weight of samples that are difficult to classify, so that the model can pay more attention to challenging samples, thereby improving the performance of the model on imbalanced datasets.

[0090] A specific embodiment is given below.

[0091] The PTB-XL ECG dataset is a large dataset containing 21,837 clinical 12-lead ECGs from 18,885 patients, each recording lasting 10 seconds. The raw waveform data was annotated by up to two cardiologists, who assigned multiple possible ECG statements to each record. The dataset includes 71 ECG statements that meet the SCP-ECG standard, covering diagnosis, modality, and rhythm information.

[0092] The data details of PTB-XL are shown in Tables 1 and 2 ;

[0093] Table 1

[0094]

[0095] Table 1 shows the six diagnostic categories and their corresponding number of ECG signals in the dataset;

[0096] Table 2

[0097]

[0098] Table 2 lists the superdiagnoses and subdiagnoses;

[0099] The dataset covers a variety of clinical diseases and has high credibility.

[0100] Use the dataset to train the network and measure the performance with F1, recall and precision. The formula is as follows:

[0101]

[0102] Among them, TP stands for true positive, which means that the model correctly predicts positive instances as positive; FN stands for false negative, which means that the model incorrectly predicts positive instances as negative; FP is false positive, which means that the model incorrectly predicts negative instances as positive. Recall can be regarded as not missing any people with heart disease, while Precision can be regarded as avoiding misdiagnosis of healthy people as people with heart disease. F1 represents the balance between recall and precision. In the ECG dataset, the F1 score is selected as the evaluation metric; Figure 6 The F1 score of the network on 15 different diseases is shown, and the F1 score of 10 diseases exceeds 0.8.

[0103] In summary, this example filters the signal through the improved SSA algorithm and uses DMSANet for classification. This example not only improves the quality of ECG signals, but also improves the accuracy of classification.

[0104] The present invention also relates to a computer-readable storage medium in its application, on which is stored an electrocardiogram classification program based on improved SSA and DMSANet, and when the program is executed by a processor, the above-mentioned electrocardiogram classification method based on improved SSA and DMSANet is implemented.

[0105] The present invention also relates to a computer device in its application, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned electrocardiogram classification method based on improved SSA and DMSANet is implemented.

[0106] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention 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, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes 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 a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0108] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0110] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0111] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An electrocardiogram classification method based on improved SSA and DMSANet, characterized by: The method obtains an electrocardiogram signal, performs noise reduction processing on the electrocardiogram signal using an improved SSA algorithm; constructs a DMSANet, including a backbone network, sub-processing modules of different scales arranged in parallel, and a multi-scale feature fusion module, and classifies the noise-reduced electrocardiogram signal using the DMSANet.

2. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 1, characterized in that: The improved SSA algorithm includes a signal decomposition method, a principal component confirmation method and a signal reconstruction method which are arranged in sequence. After the electrocardiogram signal is decomposed by the signal decomposition method, the principal component confirmation method is used to obtain the effective part of the electrocardiogram signal, and the electrocardiogram signal is reconstructed by the signal reconstruction method.

3. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 2, characterized in that: The signal decomposition method comprises the following steps: S1.1.1 Let the ECG signal be a sequence of length N, and use embedding operation to map it into a vector sequence of length L, and map it into the L×K space to obtain the trajectory matrix X, 1 <L<N; S1.1.2 performs singular value decomposition on the trajectory matrix X.

4. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 2, characterized in that: The principal element confirmation method comprises the following steps: S1.2.1 construct a new index, and reconstruct the matrix of the electrocardiogram signal based on the new index; S1.2.2 mapping the reconstructed matrix into a set of reconstructed one-dimensional time series by diagonal averaging method, and defining each one-dimensional time series as a principal element of the original electrocardiogram signal; S1.2.3 Determine and select effective principal components; S1.2.4 Extract features based on effective principal components.

5. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 4 is characterized in that: S1.2.3 includes the following steps: S1.2.3.1 Take any pivot as sample data, and whether it is selected as the target pivot is used as the sample label, and the sample label is 0 or 1; S1.2.3.2 Apply the logistic regression model to the sample data, combine the Sigmod function with the linear regression function, and calculate the probability that any sample data belongs to the sample label; S1.2.3.3 Use the gradient descent algorithm and the cross-entropy loss function to extract key features from each single principal component as the input of the logistic regression model, and determine whether the current principal component is a valid principal component based on the output of the logistic regression model.

6. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 4, characterized in that: The dimensions of the features include frequency domain, nonlinearity, and time domain.

7. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 1, characterized in that: The backbone network is based on ResNet, including a combination of two groups of convolutional layers and CBAM modules, using LeakyReLU for nonlinear changes and feature extraction.

8. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 7, characterized in that: The backbone network includes a CNN backbone network, a convolutional layer and two maximum pooling layers arranged in sequence, and three groups of first convolutional layers and CBAM module combinations and four groups of second convolutional layers and CBAM module combinations are arranged in sequence between the two maximum pooling layers.

9. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 1, characterized in that: The DMSANet includes two sub-processing modules; any sub-processing module includes three groups of third convolutional layers and CBAM modules, two groups of fourth convolutional layers and CBAM modules, and a global pooling layer.

10. The electrocardiogram classification method based on improved SSA and DMSANet according to claim 1, characterized in that: Loss optimization is performed based on focal loss.

Images