CN118152914B - Semantic structure diagram guided ECG self-coding method and system - Google Patents

Abstract

The invention belongs to the field of biological signal data machine learning, and provides a semantic structure diagram guided ECG self-coding method and system, wherein the method comprises the steps of obtaining an original electrocardiogram ECG signal; constructing a similarity matrix of the original signals by using the original electrocardiogram ECG signals, and constructing a semantic similarity graph by using label information corresponding to the original electrocardiogram ECG signals; fusing a similarity matrix and a semantic similarity graph of the original signals to construct a semantic structure diagram; based on a semantic structure diagram, encoding by using a pre-trained automatic encoder to obtain an encoded ECG signal; the loss function trained by the automatic encoder introduces a semantic structure diagram and a suppression parameter matrix on the basis of the original loss function so as to supervise the learning of hidden space features and keep key information of an ECG signal. Under the condition of supervised learning, the invention carries out self-coding processing on the signals by fusing semantic information and inherent structure information of the signals and amplifies wave bands related to semantic anomalies in the signals.

Description

ECG self-encoding method and system guided by semantic structure graph

Technical Field

The present invention belongs to the technical field of machine learning of biological signal data, and in particular relates to an ECG self-encoding method and system guided by a semantic structure graph.

Background Art

The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

Medical diagnosis can identify the disease or condition suffered by patients and explain the development of their symptoms. However, with limited healthcare services and physician resources, the diagnostic process is often time-consuming and susceptible to subjective interpretation and differences between observers. Therefore, computer-assisted intervention in clinical diagnosis can improve the quality of the healthcare system and reduce costs. In recent years, with the development of deep learning, deep learning that integrates feature engineering tasks into learning tasks has provided an exciting way to address these needs and has also been widely used in biosignal recognition and detection. However, unlike media data such as images, text, and videos, biosignals have a much higher probability of recognition error than the above application scenarios. Taking electrocardiogram (ECG) signals as an example, once an error occurs in the detection, the result may have a fatal impact on the patient. Generally, the collected ECG signals are accompanied by noise such as electrode motion artifacts, muscle artifacts, and baseline drift. These noises will adversely affect the shape characteristics of the ECG signal, thus posing a challenge to feature engineering, so noise reduction processing is required. At present, most noise reduction methods are unsupervised, and the extracted features lack the original semantic information of the signal, which is not conducive to subsequent intelligent detection.

A normal cardiac electrical signal is a complex waveform depicted by an electrocardiogram oscilloscope, and one cycle usually includes a P wave, a QRS complex wave, and a T wave. When the heart beats steadily, a typical ECG signal is formed according to a specific rule. Among them, the P wave is the first peak, which reveals the conduction process of the electrical impulse between the two atria of the heart. When the atria contract, blood is quickly pumped into the ventricles, and then quickly relaxes. This step is essential to ensure the normal flow of blood. The electrical impulse is transmitted to the ventricles and recorded in the QRS complex wave. As an important part of the electrocardiogram, the QRS complex wave reflects the electrical activity of the heart. When the ventricular contraction ends, the T wave appears, indicating that the electrical impulse has stopped propagating and the ventricles have relaxed. The specific structure is shown in Figure 1. The electrocardiogram can be used to detect and observe various heart diseases and arrhythmias, which helps doctors to detect and deal with potential health problems in a timely manner, thereby providing patients with better treatment and care. Therefore, the electrocardiogram plays an important role in the diagnosis and treatment of heart disease.

There are a large number of irrelevant features in high-dimensional ECG data, which makes it difficult for unsupervised machine learning technology to achieve high sensitivity and specificity at the same time. Moreover, after the noise reduction feature extraction process, some weak band features in the ECG may be deleted because they are not obvious, which will directly affect the detection accuracy of subsequent intelligent algorithms.

Summary of the invention

In order to solve the above problems, the present invention proposes a semantic structure graph-guided ECG self-encoding method and system. Under supervised learning, the present invention performs self-encoding processing on the signal by fusing the semantic information and inherent structure information of the signal, and amplifies the bands related to semantic abnormalities. The ECG perception signal obtained after encoding can effectively promote the subsequent intelligent algorithm to perform feature distillation again, and facilitate the deep machine learning intelligent algorithm to accurately locate the target of the ECG signal, laying the foundation for accurate identification and analysis of ECG signal categories.

According to some embodiments, a first solution of the present invention provides a semantic structure diagram guided ECG self-encoding method, which adopts the following technical solution:

A semantic structure graph guided ECG self-encoding method, comprising:

Obtaining original electrocardiogram (ECG) signal;

The original electrocardiogram (ECG) signal is used to construct a similarity matrix of the original signal, and the label information corresponding to the original electrocardiogram (ECG) signal is used to construct a semantic similarity graph;

The similarity matrix and semantic similarity graph of the original signal are integrated to construct a semantic structure graph;

Based on the semantic structure graph, a pre-trained autoencoder is used for encoding to obtain an encoded ECG signal;

Among them, the loss function of the automatic encoder training introduces a semantic structure graph and an inhibition parameter matrix on the basis of the original loss function to supervise the learning of latent space features and retain the key information of the ECG signal.

Furthermore, the K nearest neighbor method is used to construct a similarity matrix of the original signal using the original electrocardiogram (ECG) signal.

Furthermore, the semantic similarity graph is constructed by using the label information corresponding to the original electrocardiogram (ECG) signal, specifically:

If two different raw ECG signals share a label, it is 1, otherwise it is zero;

If the label information corresponding to the original ECG signal is a one-hot vector, then when the product of the transpose of the label information and the label information corresponding to another original signal is greater than zero, it is 1;

A semantic similarity graph is constructed based on the above principles.

Furthermore, the similarity matrix and the semantic similarity graph of the fused original signal are used to construct a semantic structure graph, specifically:

if This means that the original signals x _i and x _j are both close neighbors and share the same label, belonging to the intra-class close neighbor relationship, and have the strongest similarity;

if This means that _xi and _xj share the same label but are not in a close neighbor relationship. They are in a non-close neighbor relationship within the class, and the similarity is not strong enough. They are original signals that need to strengthen the similarity.

if Then _xi and _xj are neighbors but do not belong to the same category. This situation belongs to the category where the similarity needs to be weakened;

if This means that the original signals _xi and _xj are neither neighbors nor belong to the same category, and the similarity is 0;

in, is an element in the semantic similarity graph, is an element in the similarity matrix of the original signal.

Furthermore, the encoding is performed using a pre-trained autoencoder to obtain an encoded ECG signal, specifically:

Encoder Utilization It means that given an original signal x _i , its potential feature z _i can be obtained by the following formula:

in, represents the k-dimensional potential representation of the original signal _xi , Represents the weights that need to be learned during the encoding process;

use The network output represented by , the decoder function is represented by f _θ , and the reconstructed It can be decoded from z _i to get:

Among them, θ represents the weight parameter that needs to be learned in the decoding process.

Furthermore, the loss function of the autoencoder training introduces a semantic structure graph and an inhibition parameter matrix based on the original loss function to supervise the learning of latent space features, specifically:

Among them, γ,λ>0 are the hyperparameters of the semantic structure correction part and the regularization part respectively. is the loss function of the semantic structure graph module, is the binary cross entropy loss function, is the regularized loss function.

Furthermore, the semantic structure graph module loss function is defined as follows:

in, ⊙ represents element product, semantic structure diagram Suppression parameter matrix Potential signal representation

The regularized loss function is specifically:

Among them, the regularization term Row Sparsity Regularization

According to some embodiments, a second solution of the present invention provides an ECG self-encoding system guided by a semantic structure graph, which adopts the following technical solution:

A semantic structure graph guided ECG self-encoding system, comprising:

A signal acquisition module is configured to acquire an original electrocardiogram (ECG) signal;

A semantic structure graph construction module is configured to construct a similarity matrix of the original signal using the original electrocardiogram ECG signal, and to construct a semantic similarity graph using label information corresponding to the original electrocardiogram ECG signal;

The similarity matrix and semantic similarity graph of the original signal are integrated to construct a semantic structure graph;

A signal autoencoding module is configured to encode based on the semantic structure graph using a pre-trained autoencoder to obtain an encoded ECG signal;

Among them, the loss function of the automatic encoder training introduces a semantic structure graph and an inhibition parameter matrix on the basis of the original loss function to supervise the learning of latent space features and retain the key information of the ECG signal.

According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium.

A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a semantic structure graph-guided ECG self-encoding method as described in the first aspect above.

According to some embodiments, a fourth aspect of the present invention provides a computer device.

A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a semantic structure diagram guided ECG self-encoding method as described in the first aspect above are implemented.

Compared with the prior art, the present invention has the following beneficial effects:

The present invention provides a semantic structure graph guided ECG self-encoding method and system. Under supervised learning, the encoding process of the electrocardiogram (ECG) signal is supervised by fusing the semantic information and inherent structure information of the signal, so that the extracted code contains similar information of the same label and neighboring signals at the same time, and weakens other information to achieve signal noise reduction effect. The ECG perception signal obtained after encoding can effectively promote the subsequent intelligent algorithm to perform feature distillation again, and facilitate the deep machine learning intelligent algorithm to accurately locate the target of the ECG signal, laying the foundation for accurate identification and analysis of ECG signal categories.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

FIG1 is a flow chart of an ECG self-encoding method guided by a semantic structure graph according to an embodiment of the present invention;

FIG2 is a schematic diagram of the definition of each band of a standard ECG signal within one cycle according to an embodiment of the present invention;

FIG3 is a diagram of an autoencoder network structure based on a semantic structure diagram in an embodiment of the present invention;

FIG. 4 is a visualized comparison diagram of the extracted coded signal and the original signal in an embodiment of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and/or combinations thereof.

In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

Embodiment 1

As shown in Figure 1, this embodiment provides an ECG self-encoding method guided by a semantic structure diagram. This embodiment uses the method applied to a server as an example. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application. In this embodiment, the method includes the following steps:

Obtaining original electrocardiogram (ECG) signal;

The original electrocardiogram (ECG) signal is used to construct a similarity matrix of the original signal, and the label information corresponding to the original electrocardiogram (ECG) signal is used to construct a semantic similarity graph;

The similarity matrix and semantic similarity graph of the original signal are integrated to construct a semantic structure graph;

Based on the semantic structure graph, a pre-trained autoencoder is used for encoding to obtain an encoded ECG signal;

Among them, the loss function of the automatic encoder training introduces a semantic structure graph and an inhibition parameter matrix on the basis of the original loss function to supervise the learning of latent space features and retain the key information of the ECG signal.

As shown in Figure 2, given an ECG signal set in, represents an ECG signal of dimension m, Represents the label information corresponding to the signal _xi .

In order to construct the ECG autoencoder, for the training data, we first construct a semantic structure graph based on the inherent structure of the original signal and the semantic similarity graph. The detailed process is as follows:

(1) First, the original signal _xi is used to construct the inherent structure graph. For simplicity, the similarity matrix of the original signal is constructed using K nearest neighbors. The definition is as follows:

(2) Using the label information of the dataset Constructing a semantic similarity graph If the original signals _xi and _xj share a label, then let Otherwise, it is assigned a value of 0. If the label _yi is a one-hot vector, then hour, The specific definitions are as follows:

(3) Fusion of inherent similarity structure matrix and semantic similarity graph Constructing semantic structure graph The detailed process is as follows: If This means that the original signals _xi and _xj are both close neighbors and share the same label, belonging to the close neighbor relationship within the class, and the similarity is the strongest; if This means that _xi and _xj share the same label but are not neighbors, which means they are not in a close proximity within the class. The similarity is not strong enough, and they are original signals that need to strengthen the similarity. Then _xi and _xj are neighbors but do not belong to the same category. This situation needs to weaken the similarity. If This means that the original signals _xi and _xj are neither neighbors nor belong to the same class, and the similarity is 0. The specific definition is as follows:

Among them, ∈>0 is the threshold parameter that controls the similarity.

Completed the semantic structure diagram You can use To constrain the extracted latent signal representation That is to keep Z as close as possible. However, due to the semantic structure graph Highly sparse, direct approximation may result in excessive loss of the potential signal Z, so the suppression parameter matrix is introduced Its definition is as follows:

In order to facilitate optimization, the semantic structure graph module loss is defined as follows:

in, ⊙ represents element-wise product.

This loss can then be used to correct the training of the autoencoder. The network structure of the autoencoder based on the semantic structure graph is shown in Figure 3. The essence is to add a semantic structure graph to the hidden layer of the autoencoder network to guide the output of the correction network.

An autoencoder is an unsupervised deep network model that learns the implicit features of input data for encoding and uses the learned new features to decode and reconstruct the original input data. This method uses the nonlinear characteristics of neural networks to effectively compress the original data, achieve feature dimensionality reduction and noise reduction, and apply the extracted new features to the supervised learning model to complete subsequent intelligent detection tasks. However, the latent space of the autoencoder lacks an interpretable and usable structure, and cannot ensure that the main structural information can be retained during the dimensionality reduction process, and may even suffer from severe overfitting. This also means that some latent space features may become meaningless after decoding.

In view of this, the semantic similarity matrix is constructed using label information to supervise the learning of latent space features, and the graph structure of the original data is introduced to maintain its inherent structure, so that the learned potential features can maintain the original similarity structure as much as possible, as shown in Figure 3. To achieve this goal, this embodiment adds formula (5) on the basis of the original autoencoder loss function. The specific formula is as follows:

Encoder Utilization It means that, given an original signal x _i , its potential feature z _i can be obtained by formula (6):

in, represents the k-dimensional potential representation of the original signal _xi , Represents the weights that need to be learned during the encoding process.

use The network output represented by , the decoder function is represented by f _θ , and the reconstructed It can be decoded from z _i to get:

Among them, θ represents the weight parameter that needs to be learned in the decoding process. The goal of the autoencoder is to reconstruct the original signal, that is, To achieve this goal, we need to define a loss function to measure the difference between the two

Two common loss functions for autoencoders are mean square error (RMSE) and binary cross entropy. The main difference between the two is that binary cross entropy has a stronger penalty for large errors and can average the reconstructed signal values. Therefore, binary cross entropy loss is used. To use this loss, this embodiment normalizes the original signal _xi to [0,1]. The binary cross entropy loss (LCE) is defined as follows:

In addition, regarding the activation function of the network, we use the sigmoid function for the input layer and the output layer, and the SERLU function for the middle layer, ie, Because the SERLU function has a self-regularization property, the output of SERLU can be statistically pushed to zero mean.

In addition, for the weight parameter Θ = {W ¹ ,W ² ,…,W ^layer }, this embodiment expects to select the most discriminative feature through the weight parameter, that is, if w _i = 0, it means that the corresponding i-th feature does not contribute much to other features, so it can be sparse, and the features with important significance are weighted. For this purpose, row sparse regularization is applied. In addition, in order to prevent the network from over-learning, a regularization term is also applied Therefore, the entire regularization loss can be defined as:

Combining formulas (5), (8) and (9) we can get our final loss function:

Among them, γ, λ>0 are the hyperparameters of the semantic structure correction part and the regularization part respectively.

Experimental verification part

In order to verify the effectiveness of the proposed semantic structure graph-based autoencoder, a public ECG dataset was used for verification. The experimental verification is divided into two tasks: clustering analysis using the extracted latent signal and recognition task using the latent signal. The k-means algorithm is used for clustering experiments and the nearest neighbor classifier is used for classification experiments to evaluate the performance of different autoencoder feature selection methods.

Dataset: The 2017 PhysioNet/CinC Challenge ECG dataset consists of 8528 ECG segments with durations ranging from 30 seconds to 60 seconds. The dataset is divided into four groups: normal sinus rhythm (N), atrial fibrillation (A), other rhythms (O), and noisy ECG (～).

In order to verify the discriminative ability of the autoencoder to extract ECG signals, 500 samples were randomly selected from 3000 samples, and the original signal and the encoded signal were visually compared, and the results are shown in Figure 4. The results in Figure 4 show that by adding a semantic structure graph to the potential signal output layer to guide learning, the signal encoding process can incorporate information about adjacent nodes, connected edges, and global information, modulate the features of this layer, and then capture the information of the attributes of interest, making the final extracted encoding information more conducive to subsequent recognition and clustering tasks.

In order to simplify the training time, 3000 sampled signals are randomly selected for verification. During the experiment, the clustering and recognition results of the original signal and the data signal encoded by the autoencoder are compared, and the accuracy rate (ACC) is used to measure the performance. The clustering ACC is defined as:

Where p _i represents the true label, q _i is the cluster label result of the original signal _xi , and map(q _i ) uses the Kuhn-Munkres algorithm to sort the cluster labels to obtain the best mapping function that matches the true value label.

The clustering and classification results are shown in Table 1. The results show that the feature engineering performed by the proposed semantic structure autoencoder can not only effectively reduce the feature dimension, but also effectively improve the performance of clustering and classification. This is due to the fact that the semantic structure autoencoder can effectively capture more discriminative features by maintaining the structural information and semantic information of the original signal. It can be seen that the proposed autoencoder uses all features with different weights to nonlinearly represent each feature. At the same time, by minimizing the reconstruction error and regularizing the semantic structure graph, a subset of observed features that retains the inherent structural information and semantic confidence of the original data is obtained.

Table 1 Experimental structure comparison

Task Original Features Autoencoder Sparse Autoencoder Semantic Structure Graph Sparse Autoencoder Clustering (ACC%±std) 44.9±2.9 45.5±0.3 45.3±1.6 47.7±6.4 Classification(ACC%) 69.9 72.1 72.2 72.7

Embodiment 2

This embodiment provides an ECG self-encoding system guided by a semantic structure graph, including:

A signal acquisition module is configured to acquire an original electrocardiogram (ECG) signal;

A semantic structure graph construction module is configured to construct an inherent similarity structure matrix using the original electrocardiogram ECG signal, and to construct a semantic similarity graph using label information corresponding to the original electrocardiogram ECG signal;

The inherent similarity structure matrix and semantic similarity graph are integrated to construct the semantic structure graph;

A signal autoencoding module is configured to encode based on the semantic structure graph using a pre-trained autoencoder to obtain an encoded ECG signal;

Among them, the loss function of the automatic encoder training introduces a semantic structure graph and an inhibition parameter matrix on the basis of the original loss function to supervise the learning of latent space features and retain the key information of the ECG signal.

The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment 1. It should be noted that the above modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

The proposed system can be implemented in other ways. For example, the system embodiment described above is only illustrative, and the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

Embodiment 3

This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the ECG self-encoding method guided by a semantic structure graph as described in the first embodiment above are implemented.

Embodiment 4

This embodiment provides a computer device, 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 steps in the ECG self-encoding method guided by a semantic structure graph as described in the first embodiment above are implemented.

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 hardware embodiments, software embodiments, or embodiments 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 and optical storage, etc.) containing computer-usable program codes.

The present invention is described with reference to the flowchart and/or block diagram of the method, device (system), and computer program product according to the embodiment of the present invention. It should be understood that each process and/or box in the flowchart and/or block diagram, as well as the combination of the process and/or box 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 produce a device for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims (7)

1. A semantic structure guided ECG self-encoding method, comprising:

acquiring an original Electrocardiogram (ECG) signal;

Constructing a similarity matrix of the original signals by using the ECG signals of the original electrocardiogram, and constructing a semantic similarity graph by using label information corresponding to the ECG signals of the original electrocardiogram, wherein the semantic similarity graph is specifically as follows:

Given a set of ECG signals Wherein, An ECG signal with dimension m is represented, and y _i epsilon N represents label information corresponding to the signal x _i;

constructing an inherent structure diagram by using an original signal x _i, and constructing a similarity matrix of the original signal by using K nearest neighbor The definition is as follows:

constructing semantic similarity graph by using label information y _i epsilon N of dataset If the original signals x _i and x _j share a tag, then letOtherwise, value 0 is assigned, if the tag y _i is a one-hot vector, then whenIn the time-course of which the first and second contact surfaces,The specific definition is as follows:

Fusing a similarity matrix and a semantic similarity graph of original signals to construct a semantic structure diagram, wherein the semantic structure diagram comprises the following concrete steps:

Fusion similarity matrix And semantic similarity graphConstruction of semantic structure diagramWherein, Is an element in the semantic similarity graph,Is an element in the similarity matrix of the original signal;

If it is Then the original signals x _i and x _j are the neighbor relations and share the same label, belong to the intra-class close-proximity relations, and have the strongest similarity; if it isThen x _i and x _j are illustrated to share the same label but not a neighbor relationship, belong to a non-close relationship in the class, have insufficient similarity, and belong to an original signal needing to strengthen the similarity; if it isX _i and x _j are close-proximity relationships but do not belong to the same class, which belongs to the category where weakening of similarity is required; if it isThen it means that the original signals x _i and x _j are neither neighbor relations nor belonging to the same class, with a similarity of 0; the specific definition is as follows:

wherein, e >0 is a threshold parameter controlling similarity; by means of To constrain the extracted potential signal representationsThus, the semantic structure diagram module loss function is defined as follows:

Wherein, As indicated by the element product, the parameter matrix is suppressedLatent signal representation

Based on a semantic structure diagram, encoding by using a pre-trained automatic encoder to obtain an encoded ECG signal;

The loss function trained by the automatic encoder introduces a semantic structure diagram and a suppression parameter matrix on the basis of the original loss function so as to supervise the learning of hidden space features and keep key information of an ECG signal.

2. The semantic structure guided ECG self-coding method of claim 1, wherein the encoding with a pre-trained automatic encoder yields an encoded ECG signal, specifically:

Encoder utilization Representing, given an original signal x _i, its potential characteristic z _i is derived from the following equation:

Wherein, A k-dimensional potential representation of the original signal x _i,A right parameter which indicates that the coding process needs to learn;

By using The network output of the representation, the decoder function is denoted by f _θ, the reconstructedDecoding from z _i:

Where θ represents the rights that the decoding process needs to learn.

3. The semantic structure guided ECG self-coding method of claim 1, wherein the automatic encoder trained loss function introduces semantic structure and suppression parameter matrix based on original loss function to supervise learning of hidden space features, specifically:

Wherein, gamma, lambda >0 are the super-parameters of the semantic structure correcting part and the regularization term part respectively, Is a semantic structure diagram module loss function,Is a binary cross entropy loss function,Is a regularized loss function.

4. A semantic structure guided ECG self-encoding method according to claim 3, characterized in that the regularization loss function is:

Wherein regularization term Line sparse regularization

5. A semantic structure guided ECG self-encoding system, comprising:

a signal acquisition module configured to acquire an original electrocardiogram, ECG, signal;

the semantic structure diagram construction module is configured to construct a similarity matrix of the original signals by using the original electrocardiogram ECG signals, and construct a semantic similarity diagram by using label information corresponding to the original electrocardiogram ECG signals, and specifically comprises the following steps:

Given a set of ECG signals Wherein, An ECG signal with dimension m is represented, and y _i epsilon N represents label information corresponding to the signal x _i;

constructing an inherent structure diagram by using an original signal x _i, and constructing a similarity matrix of the original signal by using K nearest neighbor The definition is as follows:

constructing semantic similarity graph by using label information y _i epsilon N of dataset If the original signals x _i and x _j share a tag, then letOtherwise, value 0 is assigned, if the tag y _i is a one-hot vector, then whenIn the time-course of which the first and second contact surfaces,The specific definition is as follows:

Fusing a similarity matrix and a semantic similarity graph of original signals to construct a semantic structure diagram, wherein the semantic structure diagram comprises the following concrete steps:

Fusion similarity matrix And semantic similarity graphConstruction of semantic structure diagramWherein, Is an element in the semantic similarity graph,Is an element in the similarity matrix of the original signal;

If it is Then the original signals x _i and x _j are the neighbor relations and share the same label, belong to the intra-class close-proximity relations, and have the strongest similarity; if it isThen x _i and x _j are illustrated to share the same label but not a neighbor relationship, belong to a non-close relationship in the class, have insufficient similarity, and belong to an original signal needing to strengthen the similarity; if it isX _i and x _j are close-proximity relationships but do not belong to the same class, which belongs to the category where weakening of similarity is required; if it isThen it means that the original signals x _i and x _j are neither neighbor relations nor belonging to the same class, with a similarity of 0; the specific definition is as follows:

wherein, e >0 is a threshold parameter controlling similarity; by means of To constrain the extracted potential signal representationsThus, the semantic structure diagram module loss function is defined as follows:

Wherein, As indicated by the element product, the parameter matrix is suppressedLatent signal representation

The signal self-coding module is configured to code by utilizing a pre-trained automatic coder based on a semantic structure diagram to obtain a coded ECG signal;

The loss function trained by the automatic encoder introduces a semantic structure diagram and a suppression parameter matrix on the basis of the original loss function so as to supervise the learning of hidden space features and keep key information of an ECG signal.

6. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of a semantic structure guided ECG self-encoding method according to any one of claims 1-4.

7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of a semantic structure guided ECG self-encoding method according to any one of claims 1-4 when the program is executed.