Electrocardiogram signal classification method, system and device based on wavelet scattering transform

Abstract

The application discloses a kind of electrocardiosignal classification method, system and equipment based on wavelet scattering transform.The method comprises the following steps: step 1, the R peak interval feature vector of each heartbeat signal in the electrocardiosignal obtained by calculation;Step 2, the electrocardiosignal is segmented into a group of heartbeat signals with single heartbeat as unit;Step 3, the heartbeat signal is carried out wavelet scattering transform, and the scattering spectrum of heartbeat signal is obtained;Step 4, the scattering spectrum is input into feature encoder, and the morphological feature vector of heartbeat signal is obtained;Step 5, the R peak interval feature vector and morphological feature vector are input into feature fusion classifier, and the category of the heartbeat signal is obtained.The method of the application can extract the representative time-frequency features of different categories of electrocardiosignal from small sample electrocardiosignal data, to realize the accurate classification of electrocardiosignal.

Description

Technical Field

[0001] This invention relates to the technical field of electrocardiogram (ECG) signal analysis, specifically to an ECG signal classification method, system, and device based on wavelet scattering transform. Background Technology

[0002] Electrocardiogram (ECG) signals are electrical signals collected through metal electrodes placed on a patient's skin. They reflect the patient's heart activity in real time and provide important reference information in clinical diagnosis. However, some medical scenarios require long-term monitoring of a patient's physiological state. The resulting ECG data volume is large, and manual analysis by professional physicians is time-consuming, hindering timely and accurate clinical diagnosis. Therefore, numerous automated ECG signal processing methods have emerged. These methods generally include preprocessing, feature extraction, and classification. In traditional ECG signal classification methods, preprocessing and feature extraction rely on specific signal processing methods and domain-specific expertise, resulting in a failure to fully represent the underlying characteristics of ECG signals.

[0003] In recent years, with the improvement of hardware computing power, the excellent expressive ability of deep learning technology for nonlinear mappings has been widely verified in fields such as image recognition and natural language processing. Naturally, it has also been introduced into the field of electrocardiogram (ECG) signal analysis, leading to the proposal of various end-to-end ECG signal classification methods. Unlike traditional classification methods, these methods do not require preprocessing or feature extraction of the ECG signal; the ECG signal is directly input into a classifier composed of neural networks, which then outputs the corresponding ECG signal category. However, neural networks have a large number of parameters, and the accuracy of these parameters depends on the training and feedback adjustments of training samples. This means that neural network-based classifiers require a large amount of training data and labels to achieve good generalization ability.

[0004] Inspired by the architecture of convolutional neural networks (CNNs), Mallat proposed the wavelet scattering transform. The algorithmic structure of the wavelet scattering transform is similar to that of CNNs, also including convolution, nonlinear operations, and pooling. Unlike CNNs, the convolution kernel of the wavelet scattering transform is a fixed wavelet filter, which does not require training. The signal undergoes a semi-discrete wavelet transform, followed by a nonlinear modulo operation, and finally filtering with a low-pass filter to obtain the scattering spectrum representing the abstract time-frequency features of the signal. The obtained scattering spectrum can be iteratively subjected to wavelet scattering transforms to obtain higher-dimensional features. Rigorous mathematical theory proves that the features obtained through the wavelet scattering transform possess translation invariance and deformation stability. This gives the wavelet scattering transform certain advantages in signal feature extraction tasks. Summary of the Invention

[0005] The purpose of this invention is to address the problems of poor accuracy and robustness in existing electrocardiogram (ECG) signal classification techniques and to provide an ECG signal classification method based on wavelet scattering transform. This method can extract representative time-frequency features of different categories of ECG signals from small sample ECG data, achieving accurate classification of ECG signals and overcoming the limitation of deep learning-based methods requiring a large amount of training data.

[0006] The specific technical solution adopted in this invention is as follows:

[0007] In a first aspect, the present invention provides a method for classifying electrocardiogram signals based on wavelet scattering transform, which includes the following steps:

[0008] S1. Calculate the R-peak interval feature vector of each heartbeat signal in the acquired electrocardiogram signal;

[0009] S2. Divide the electrocardiogram signal into a group of heartbeat signals, with each heartbeat as a unit;

[0010] S3. Perform wavelet scattering transform on the heartbeat signal obtained in S2 to obtain the scattering spectrum of the heartbeat signal;

[0011] S4. Input the scattering spectrum obtained in S3 into the feature encoder to obtain the morphological feature vector of the heartbeat signal;

[0012] S5, input the R-peak interval feature vector obtained in S1 and the morphological feature vector obtained in S4 into the feature fusion classifier to obtain the category of the heartbeat signal.

[0013] Preferably, the electrocardiogram signal in S1 is an ECG signal, which is acquired by electrodes placed on the patient's body surface and used to record changes in electrical activity caused by cardiac activity.

[0014] As a preferred embodiment, the specific implementation method of S1 is as follows:

[0015] S101. Using the Pan-Tompkin algorithm, detect the time points where the R peaks of each heartbeat occur in the electrocardiogram signal;

[0016] S102. Subtract the time point of the R peak of each heartbeat detected in S101 from the time point of the R peak of the previous heartbeat to obtain the interval time between each heartbeat and the previous heartbeat as the previous interval time.

[0017] S103. Subtract the time point of the R peak of each heartbeat detected in S101 from the time point of the R peak of the next heartbeat to obtain the interval time between each heartbeat and the next heartbeat as the next interval time.

[0018] S104. Normalize the pre-interval time and post-interval time of each heartbeat obtained in S102 and S103 to obtain the normalized pre-interval time and normalized post-interval time.

[0019] S105. Subtract the normalized interval time from the normalized interval time of each heartbeat obtained in S104 to obtain the interval time difference of each heartbeat.

[0020] S106. Combine the pre-normalized interval time, post-normalized interval time, and interval time difference obtained in S104 and S105 to form the R-peak interval feature vector.

[0021] As a preferred embodiment, the specific implementation method of S2 is as follows:

[0022] S201. Using the Pan-Tompkin algorithm, detect the time points of the R peaks of each heartbeat in the electrocardiogram signal;

[0023] S202. Based on the time point of the R peak obtained in S201, calculate the midpoint between two adjacent R peaks in the electrocardiogram signal;

[0024] S203. The midpoint between the R peaks obtained in S202 is used as the endpoint of the heartbeat signal. The ECG signal is segmented using these endpoints to obtain a set of heartbeat signals.

[0025] S204. For each heartbeat signal obtained in S203, if the time interval between the R peak and the signal endpoint of the heartbeat signal is less than 0.4 seconds, then the heartbeat signal is padded with zeros until the time interval between the R peak and the endpoint is 0.4 seconds; if the time interval between the R peak and the signal endpoint of the heartbeat signal is greater than 0.4 seconds, then the portion of the heartbeat signal exceeding 0.4 seconds is truncated.

[0026] Preferably, the specific process of S3 is as follows:

[0027] S301. Convolve the heartbeat signal obtained in S2 with a low-pass filter to obtain the zero-order scattering parameter, the expression of which is as follows:

[0028] S (0) x(t)=x★φ J

[0029] Where x represents the heartbeat signal, φ J S represents a low-pass filter with a center frequency of zero. (0) x represents the zeroth-order scattering parameter of signal x, and ★ represents the convolution operation;

[0030] S302. Convolve the heartbeat signal with the analysis wavelet and take the modulus to obtain the first-order wavelet parameters. Then, convolve the first-order wavelet parameters with the low-pass filter again to obtain the first-order scattering parameters, the expression of which is as follows:

[0031]

[0032] in, S represents the analytical wavelet signal with center frequency λ. (1) x represents the first-order scattering parameter of signal x;

[0033] S303. Convolve the first-order wavelet parameters obtained in S302 with the analytical wavelet and take the modulus to obtain the second-order wavelet parameters. Then convolve the second-order wavelet parameters with the low-pass filter to obtain the second-order scattering parameters, as shown in the following expression:

[0034]

[0035] in, S represents an analytical wavelet signal with a center frequency of μ. (2) x represents the second-order scattering parameter of signal x;

[0036] S304. The scattering spectrum of the heartbeat signal is composed of the first-order scattering parameters and the second-order scattering parameters of the signal x.

[0037] Preferably, the feature encoder in S4 is a fully connected neural network.

[0038] Preferably, the fully connected neural network serving as the feature encoder contains one hidden layer and one output layer, with each hidden layer and output layer containing 32 nodes, and a linear rectified unit is selected as the activation function.

[0039] Preferably, the feature fusion classifier in S5 is a fully connected neural network.

[0040] Preferably, the fully connected neural network used as the feature fusion classifier contains one hidden layer and one output layer. The hidden layer contains eight nodes and selects ReLU as the activation function, while the output layer contains five nodes and selects the normalized exponential function as the activation function.

[0041] Preferably, both the feature encoder and the feature fusion classifier are pre-trained, and the optimal model parameters obtained from the training are used for actual inference.

[0042] Secondly, the present invention provides an electrocardiogram signal classification system based on wavelet scattering transform, comprising:

[0043] The R-peak interval feature vector calculation module is used to calculate the R-peak interval feature vector of each heartbeat signal in the acquired electrocardiogram signal.

[0044] The heartbeat signal segmentation module is used to segment the electrocardiogram signal into a group of heartbeat signals based on a single heartbeat.

[0045] The scattering spectrum processing module is used to perform wavelet scattering transform on the heartbeat signal obtained by the heartbeat signal segmentation module to obtain the scattering spectrum of the heartbeat signal;

[0046] The morphological feature vector calculation module is used to input the scattering spectrum obtained by the scattering spectrum processing module into the feature encoder to obtain the morphological feature vector of the heartbeat signal.

[0047] The heartbeat signal classification module is used to input the R-peak interval feature vector obtained by the R-peak interval feature vector calculation module and the morphological feature vector obtained by the morphological feature vector calculation module into the feature fusion classifier to obtain the category of the heartbeat signal.

[0048] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the electrocardiogram signal classification method based on wavelet scattering transform as described in any of the solutions of the first aspect above.

[0049] Fourthly, the present invention provides a computer electronic device, which includes a memory and a processor;

[0050] The memory is used to store computer programs;

[0051] The processor is configured to, when executing the computer program, implement the electrocardiogram signal classification method based on wavelet scattering transform as described in any of the first aspects above.

[0052] Compared with the prior art, the beneficial effects of the present invention include:

[0053] (1) This invention does not require a large amount of data and labels for training. It can complete the feature extraction and classification of various ECG signals using small sample ECG signal data.

[0054] (2) When there is noise in the electrocardiogram signal, or when the electrocardiogram signal has a certain deformation or translation, the method of the present invention can still extract representative features in the electrocardiogram signal for classification, and has excellent robustness. Attached Figure Description

[0055] Figure 1 A plot of an electrocardiogram signal segment collected in the example is shown.

[0056] Figure 2 The scattering spectrum obtained by using the method of the present invention in the embodiment is shown.

[0057] Figure 3 This is a flowchart of the electrocardiogram signal classification method in an embodiment of the present invention.

[0058] Figure 4 This is a block diagram of the electrocardiogram signal classification system module based on wavelet scattering transform in an embodiment of the present invention. Detailed Implementation

[0059] The following section uses the ECG signal classification of the MIT-BIH arrhythmia database as an example to illustrate the implementation scheme and effect of this ECG signal classification method based on wavelet scattering transform.

[0060] In this embodiment, the MIT-BIH arrhythmia database contains 48 ECG records, each 30 minutes long, containing over 100,000 heartbeats. According to the standards proposed by the Association for the Advancement of Medical Instrumentation (AAMI), each heartbeat is classified into five categories: N (non-ectopic heartbeat), S (supraventricular ectopic heartbeat), V (ventricular ectopic heartbeat), F (mixed heartbeat), and Q (unidentified).

[0061] The electrocardiogram signals used in the embodiments are as follows: Figure 1 As shown. Figure 1 The horizontal axis represents the sampling point number, with the unit being the sample point (one sample point corresponds to the sampling interval of one data point), and the vertical axis represents the ECG signal amplitude, with the unit being millivolts (mV).

[0062] like Figure 3 As shown, this embodiment specifically employs an electrocardiogram (ECG) signal classification method based on wavelet scattering transform, which includes the following steps:

[0063] Step 1: Calculate the R-peak interval feature vector of each heartbeat signal in the acquired electrocardiogram signal.

[0064] Among them, the electrocardiogram signal is the ECG signal, which is acquired by electrodes placed on the patient's body surface and is used to record the changes in electrical activity caused by cardiac activity.

[0065] Step 1-1: Use the Pan-Tompkin algorithm to detect the time points of the R peaks of each heartbeat in the electrocardiogram signal;

[0066] Step 1-2: Subtract the time point of the R peak of each heartbeat detected in Step 1-1 from the time point of the R peak of the previous heartbeat to obtain the interval time between each heartbeat and the previous heartbeat, i.e. the previous interval time.

[0067] Steps 1-3: Subtract the time point of the R peak of each heartbeat detected in Step 1-1 from the time point of the R peak of the next heartbeat to obtain the interval time between each heartbeat and the next heartbeat, i.e. the interval time.

[0068] Steps 1-4: Normalize the pre-interval time and post-interval time of each heartbeat obtained in steps 1-2 and 1-3 to obtain the normalized heartbeat interval time, that is, the normalized pre-interval time and normalized post-interval time.

[0069] Steps 1-5: Subtract the normalized interval time from the normalized interval time of each heartbeat obtained in Steps 1-4 to obtain the interval time difference of each heartbeat.

[0070] Steps 1-6 combine the pre-normalized interval time, post-normalized interval time, and interval time difference obtained in steps 1-4 and 1-5 to form the R-peak interval feature vector.

[0071] Step 2: Divide the acquired electrocardiogram signal into a group of heartbeat signals, with each heartbeat as a unit;

[0072] Step 2-1: Use the Pan-Tompkin algorithm to detect the time points of the R peaks of each heartbeat in the electrocardiogram signal;

[0073] Step 2-2: Based on the time point of the R peak obtained in Step 2-1, calculate the midpoint between two adjacent R peaks in the electrocardiogram signal;

[0074] Step 2-3: Take the midpoint between two adjacent R peaks obtained in Step 2-2 as the endpoint of each heartbeat signal, and use these endpoints to segment the ECG signal to obtain a set of heartbeat signals.

[0075] Step 2-4: For each heartbeat signal obtained in Step 2-3, the length is unified. If the time interval between the R peak and the signal endpoint of the heartbeat signal is less than 0.4 seconds, the heartbeat signal is padded with zeros until the time interval between the R peak and the endpoint is 0.4 seconds. If the time interval between the R peak and the signal endpoint of the heartbeat signal is greater than 0.4 seconds, the portion of the heartbeat signal exceeding 0.4 seconds is truncated.

[0076] Step 3: Perform wavelet scattering transform on the heartbeat signal obtained in Step 2 to obtain the scattering spectrum of the heartbeat signal.

[0077] Wavelet scattering transform requires setting two main parameters: the invariance scale and the quality factor. The invariance scale sets the time support length of the low-pass filter, while the quality factor sets the number of wavelet filters per octave in each order of wavelet scattering transform.

[0078] In this embodiment, the time support length of the low-pass filter is set to 32 sample points. Each octave of the first-order wavelet convolution contains 8 wavelet filters, and each octave of the second-order wavelet convolution contains 1 wavelet filter.

[0079] Step 3-1: Convolve the heartbeat signal obtained in Step 2 with a low-pass filter to obtain the zero-order scattering parameters. The expression for the entire process in this step is as follows:

[0080] S (0) x(t)=x★φ J

[0081] Where x represents the heartbeat signal, φ J S represents a low-pass filter with a center frequency of zero. (0) x represents the zeroth-order scattering parameter of signal x, and ★ represents the convolution operation.

[0082] Step 3-2: Convolve the heartbeat signal obtained in Step 2 with the analytical wavelet and take the modulus to obtain the first-order wavelet parameters. Then, convolve the first-order wavelet parameters with the low-pass filter again to obtain the first-order scattering parameters. The expression for the entire process in this step is as follows:

[0083]

[0084] in, S represents the analytical wavelet signal with center frequency λ. (1) x represents the first-order scattering parameter of signal x.

[0085] Step 3-3: Convolve the first-order wavelet parameters obtained in Step 2 with the analytical wavelet and take the modulus to obtain the second-order wavelet parameters. Then, convolve the second-order wavelet parameters with the low-pass filter to obtain the second-order scattering parameters. The expression for the entire process in this step is as follows:

[0086]

[0087] in, S represents an analytical wavelet signal with a center frequency of μ. (2) x represents the second-order scattering parameter of signal x.

[0088] The scattering spectrum obtained after wavelet scattering transform of the electrocardiogram signal is as follows: Figure 2 As shown. The upper and lower parts respectively represent... Figure 1 The first-order and second-order scattering spectra of the ECG signal are shown. The horizontal axis represents the number of time windows for wavelet scattering transform, and the vertical axis represents the number of wavelet filters used in wavelet scattering transform. The center frequency of the filters decreases exponentially as the number of filters increases.

[0089] Ultimately, the scattering spectrum of the heartbeat signal is composed of the first-order and second-order scattering parameters of the signal x.

[0090] Step 4: Input the scattering spectrum obtained in Step 3 into the feature encoder to obtain the morphological feature vector of the heartbeat signal.

[0091] In this embodiment, the feature encoder is a fully connected neural network, consisting of one hidden layer and one output layer. Both the hidden and output layers contain 32 nodes, and a rectified linear unit (ReLU) is selected as the activation function. This feature encoder yields a heartbeat morphological feature vector of length 32.

[0092] Step 5: Input the R-peak interval feature vector obtained in Step 1 and the morphological feature vector obtained in Step 4 into the feature fusion classifier to obtain the category of the heartbeat signal.

[0093] In this embodiment, the feature fusion classifier is a fully connected neural network, consisting of one hidden layer and one output layer. The hidden layer contains 8 nodes and uses ReLU as the activation function; the output layer contains 5 nodes and uses the normalized exponential function (softmax) as the activation function.

[0094] In this embodiment, the MIT-BIH arrhythmia database is divided into training and testing sets according to an inter-patient protocol. The model is trained and validated using the training set, and the optimal model parameters are tested using the testing set. According to the AAMI standard, heartbeat classification methods should focus on the classification performance of S-class and V-class heartbeats. The statistical indicators of the classification results of this invention on the testing set are shown in Table 1. It can be seen that this invention has superior classification indicators for S-class and V-class heartbeats, demonstrating good classification performance.

[0095] Table 1 Statistical Indicators of Test Set Classification Results

[0096]

[0097] It should be noted that the above classification results can be used to help determine the type of heartbeat, but are not directly used for the diagnosis and treatment of diseases. The classification results can be used for non-disease diagnosis purposes, such as scientific research and testing.

[0098] Based on the same inventive concept, another preferred embodiment of the present invention also provides an electrocardiogram (ECG) signal classification system based on wavelet scattering transform, corresponding to the ECG signal classification method based on wavelet scattering transform provided in the above embodiments. For example... Figure 4 As shown, this ECG signal classification system based on wavelet scattering transform includes five basic modules, namely:

[0099] The R-peak interval feature vector calculation module is used to calculate the R-peak interval feature vector of each heartbeat signal in the acquired electrocardiogram signal.

[0100] The heartbeat signal segmentation module is used to segment the electrocardiogram signal into a group of heartbeat signals based on a single heartbeat.

[0101] The scattering spectrum processing module is used to perform wavelet scattering transform on the heartbeat signal obtained by the heartbeat signal segmentation module to obtain the scattering spectrum of the heartbeat signal;

[0102] The morphological feature vector calculation module is used to input the scattering spectrum obtained by the scattering spectrum processing module into the feature encoder to obtain the morphological feature vector of the heartbeat signal.

[0103] The heartbeat signal classification module is used to input the R-peak interval feature vector obtained by the R-peak interval feature vector calculation module and the morphological feature vector obtained by the morphological feature vector calculation module into the feature fusion classifier to obtain the category of the heartbeat signal.

[0104] Since the principle of the ECG signal classification system based on wavelet scattering transform in this embodiment of the invention is similar to the ECG signal classification method based on wavelet scattering transform in the above embodiment of the invention, any parts of the specific implementation of each module of the system in this embodiment that are not exhaustive can also refer to the specific implementation of the above method, and repeated parts will not be described again.

[0105] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the electrocardiogram signal classification method based on wavelet scattering transform provided in the above embodiments, which includes a memory and a processor;

[0106] The memory is used to store computer programs;

[0107] The processor is configured to implement the electrocardiogram signal classification method based on wavelet scattering transform as described above when executing the computer program.

[0108] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0109] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the electrocardiogram signal classification method based on wavelet scattering transform provided in the above embodiments. The storage medium stores a computer program, which, when executed by a processor, can realize the electrocardiogram signal classification method based on wavelet scattering transform as described above.

[0110] Specifically, in the computer-readable storage medium of the two embodiments described above, the stored computer program is executed by a processor, which can perform the aforementioned steps S1 to S5.

[0111] It is understood that the aforementioned storage media may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage media may also be various media capable of storing program code, such as USB flash drives, external hard drives, magnetic disks, or optical discs.

[0112] It is understood that the processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0113] It should also be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided in this application, the division of steps or modules in the system and method is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or steps may be combined or integrated together, and a module or step may also be split.

[0114] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A method for classifying electrocardiogram (ECG) signals based on wavelet scattering transform, characterized in that, Includes the following steps: S1. Calculate the R-peak interval feature vector of each heartbeat signal in the acquired electrocardiogram signal; The specific implementation method of S1 is as follows: S101. Using the Pan-Tompkin algorithm, detect the time points where the R peaks of each heartbeat occur in the electrocardiogram signal; S102. Subtract the time point of the R peak of each heartbeat detected in S101 from the time point of the R peak of the previous heartbeat to obtain the interval time between each heartbeat and the previous heartbeat as the previous interval time. S103. Subtract the time point of the R peak of each heartbeat detected in S101 from the time point of the R peak of the next heartbeat to obtain the interval time between each heartbeat and the next heartbeat as the next interval time. S104. Normalize the pre-interval time and post-interval time of each heartbeat obtained in S102 and S103 to obtain the normalized pre-interval time and normalized post-interval time. S105. Subtract the normalized interval time from the normalized interval time of each heartbeat obtained in S104 to obtain the interval time difference of each heartbeat. S106. Combine the pre-normalized interval time, post-normalized interval time, and interval time difference obtained in S104 and S105 to form the R-peak interval feature vector. S2. Divide the electrocardiogram signal into a group of heartbeat signals, with each heartbeat as a unit; S3. Perform wavelet scattering transform on the heartbeat signal obtained in S2 to obtain the scattering spectrum of the heartbeat signal; The specific process of S3 is as follows: S301. Convolve the heartbeat signal obtained in S2 with a low-pass filter to obtain the zero-order scattering parameter, the expression of which is as follows: in, represents the heart beat signal, This indicates a low-pass filter with a center frequency of zero. Indicates signal The zero-order scattering parameters, This represents the convolution operation; S302. Convolve the heartbeat signal with the analysis wavelet and take the modulus to obtain the first-order wavelet parameters. Then, convolve the first-order wavelet parameters with the low-pass filter again to obtain the first-order scattering parameters, the expression of which is as follows: in, The center frequency is Analysis of wavelet signals, Indicates signal The first-order scattering parameter; S303. Convolve the first-order wavelet parameters obtained in S302 with the analytical wavelet and take the modulus to obtain the second-order wavelet parameters. Then convolve the second-order wavelet parameters with the low-pass filter to obtain the second-order scattering parameters, as shown in the following expression: in, The center frequency is Analysis of wavelet signals, Indicates signal The second-order scattering parameters; S304, from signal The first-order scattering parameters and the second-order scattering parameters constitute the scattering spectrum of the heartbeat signal; S4. Input the scattering spectrum obtained in S3 into the feature encoder to obtain the morphological feature vector of the heartbeat signal; the feature encoder is a fully connected neural network; S5, input the R-peak interval feature vector obtained in S1 and the morphological feature vector obtained in S4 into the feature fusion classifier to obtain the category of the heartbeat signal; the feature fusion classifier is a fully connected neural network.

2. The ECG signal classification method based on wavelet scattering transform as described in claim 1, characterized in that, The electrocardiogram signal in S1 is an ECG signal, which is acquired by electrodes placed on the patient's body surface and used to record changes in electrical activity caused by cardiac activity.

3. The electrocardiogram signal classification method based on wavelet scattering transform according to claim 1, characterized in that, The specific implementation method of S2 is as follows: S201. Using the Pan-Tompkin algorithm, detect the time points of the R peaks of each heartbeat in the electrocardiogram signal; S202. Based on the time point of the R peak obtained in S201, calculate the midpoint between two adjacent R peaks in the electrocardiogram signal; S203. The midpoint between the R peaks obtained in S202 is used as the endpoint of the heartbeat signal. The ECG signal is segmented using these endpoints to obtain a set of heartbeat signals. S204. For each heartbeat signal obtained in S203, if the time interval between the R peak and the signal endpoint of the heartbeat signal is less than 0.4 seconds, then the heartbeat signal is padded with zeros until the time interval between the R peak and the endpoint is 0.4 seconds; if the time interval between the R peak and the signal endpoint of the heartbeat signal is greater than 0.4 seconds, then the portion of the heartbeat signal exceeding 0.4 seconds is truncated.

4. A classification system for electrocardiogram signals based on wavelet scattering transform, characterized in that, include: The R-peak interval feature vector calculation module is used to calculate the R-peak interval feature vector for each heartbeat signal in the acquired electrocardiogram signal; the specific method for obtaining the R-peak interval feature vector is as follows: S101. Using the Pan-Tompkin algorithm, detect the time points where the R peaks of each heartbeat occur in the electrocardiogram signal; S102. Subtract the time point of the R peak of each heartbeat detected in S101 from the time point of the R peak of the previous heartbeat to obtain the interval time between each heartbeat and the previous heartbeat as the previous interval time. S103. Subtract the time point of the R peak of each heartbeat detected in S101 from the time point of the R peak of the next heartbeat to obtain the interval time between each heartbeat and the next heartbeat as the next interval time. S104. Normalize the pre-interval time and post-interval time of each heartbeat obtained in S102 and S103 to obtain the normalized pre-interval time and normalized post-interval time. S105. Subtract the normalized interval time from the normalized interval time of each heartbeat obtained in S104 to obtain the interval time difference of each heartbeat. S106. Combine the pre-normalized interval time, post-normalized interval time, and interval time difference obtained in S104 and S105 to form the R-peak interval feature vector. The heartbeat signal segmentation module is used to segment the electrocardiogram signal into a group of heartbeat signals based on a single heartbeat. The scattering spectrum processing module is used to perform wavelet scattering transform on the heartbeat signal obtained by the heartbeat signal segmentation module to obtain the scattering spectrum of the heartbeat signal; The process of obtaining the scattering spectrum of the heartbeat signal is as follows: S301. Convolve the heartbeat signal obtained in S2 with a low-pass filter to obtain the zero-order scattering parameter, the expression of which is as follows: in, represents the heart beat signal, This indicates a low-pass filter with a center frequency of zero. Indicates signal The zero-order scattering parameters, This represents the convolution operation; S302. Convolve the heartbeat signal with the analysis wavelet and take the modulus to obtain the first-order wavelet parameters. Then, convolve the first-order wavelet parameters with the low-pass filter again to obtain the first-order scattering parameters, the expression of which is as follows: in, The center frequency is Analysis of wavelet signals, Indicates signal The first-order scattering parameter; S303. Convolve the first-order wavelet parameters obtained in S302 with the analytical wavelet and take the modulus to obtain the second-order wavelet parameters. Then convolve the second-order wavelet parameters with the low-pass filter to obtain the second-order scattering parameters, as shown in the following expression: in, The center frequency is Analysis of wavelet signals, Indicates signal The second-order scattering parameters; S304, from signal The first-order scattering parameters and the second-order scattering parameters constitute the scattering spectrum of the heartbeat signal; The morphological feature vector calculation module is used to input the scattering spectrum obtained by the scattering spectrum processing module into the feature encoder to obtain the morphological feature vector of the heartbeat signal; the feature encoder is a fully connected neural network. The heartbeat signal classification module is used to input the R-peak interval feature vector obtained by the R-peak interval feature vector calculation module and the morphological feature vector obtained by the morphological feature vector calculation module into the feature fusion classifier to obtain the category of the heartbeat signal; the feature fusion classifier is a fully connected neural network.

5. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the electrocardiogram signal classification method based on wavelet scattering transform as described in any one of claims 1 to 3.

6. A computer electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the electrocardiogram signal classification method based on wavelet scattering transform as described in any one of claims 1 to 3 when executing the computer program.

Images