A method and system for classifying and recognizing electrocardiosignal based on spatiotemporal feature fusion

Abstract

The application discloses a kind of based on spatiotemporal feature fusion electrocardiosignal classification identification method and system, it is related to biological signal classification identification technical field.The method includes: obtaining individual electrocardiosignal, extracts the spatial variability characteristic curve and time variability characteristic curve of electrocardiosignal;Three-dimensional state space is constructed, based on the spatial variability characteristic curve and time variability characteristic curve of electrocardiosignal, extracts electrocardiosignal spatiotemporal feature;Based on electrocardiosignal spatiotemporal feature, extracts the first intrinsic dynamics characteristic of electrocardiosignal;Based on the difference between the first intrinsic dynamics characteristic and the second intrinsic dynamics characteristic stored in the preset electrocardiosignal mode library, obtain the first identification error;According to the first identification error, using the preset classifier scheme of adaptation to the electrocardiosignal spatiotemporal feature is identified, and identity recognition result is obtained.Compared with prior art, the present application has high recognition accuracy and success rate, low feature dimension, small computing resource demand and fast computing speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological signal classification and recognition, and more particularly to an electrocardiosignal classification and recognition method and system based on spatiotemporal feature fusion. BACKGROUND

[0002] An electrocardiosignal is a kind of weak bioelectric signal. The classification result of the data collected from the electrocardiosignal is important auxiliary means and reference information for doctors to diagnose heart disease, and can be used for auxiliary diagnosis of various heart abnormalities, prediction of morbidity and mortality of cardiovascular diseases, etc. In addition, compared with other physiological feature-based biological recognition technology, the electrocardiosignal used as a biological recognition feature for individual identity recognition has strong anti-counterfeiting performance and can effectively guarantee information security.

[0003] In the field of identity recognition, the prior art generally researches from the perspective of the static characteristics of the electrocardiosignal, and does not deeply research the internal dynamic characteristics of the electrocardiosignal, resulting in a low accuracy of the identity recognition classification. SUMMARY

[0004] To overcome the defect of low accuracy of the electrocardiosignal classification and recognition result in the prior art, the present application provides an electrocardiosignal classification and recognition method and system based on spatiotemporal feature fusion.

[0005] To solve the above technical problems, the technical solution of the present application is as follows:

[0006] In a first aspect, an electrocardiosignal classification and recognition method based on spatiotemporal feature fusion comprises:

[0007] obtaining an individual electrocardiosignal, extracting a spatial variability feature curve and a temporal variability feature curve of the electrocardiosignal;

[0008] constructing a three-dimensional state space, extracting electrocardiospatial-temporal features based on the spatial variability feature curve and the temporal variability feature curve of the electrocardiosignal;

[0009] extracting first internal dynamic characteristics of the electrocardiosignal based on the electrocardiospatial-temporal features;

[0010] obtaining a first recognition error based on the difference between the first internal dynamic characteristics and second internal dynamic characteristics stored in a preset electrocardiosignal mode library; wherein the preset electrocardiosignal mode library is constructed according to the heart signals of different individual identities under different preset conditions;

[0011] adopting an adaptive preset classifier scheme to recognize the electrocardiospatial-temporal features according to the first recognition error, and obtaining an identity recognition result.

[0012] In a second aspect, an electrocardiosignal classification and recognition system based on spatiotemporal feature fusion comprises:

[0013] The collection module is configured to collect an electrocardiosignal of an individual.

[0014] The feature extraction module is configured to extract a spatial variability feature curve and a temporal variability feature curve of the electrocardiosignal, extract an electrocardio spatio-temporal feature based on the spatial variability feature curve and the temporal variability feature curve of the electrocardiosignal, and extract a first intrinsic dynamic feature of the electrocardiosignal based on the electrocardio spatio-temporal feature.

[0015] The identity recognition module is configured to carry a preset electrocardio mode library, the preset electrocardio mode library stores a second intrinsic dynamic feature, perform difference calculation on the first intrinsic dynamic feature and the second intrinsic dynamic feature to obtain a first recognition error, and recognize the electrocardio spatio-temporal feature based on the first recognition error by using an adaptive preset classifier scheme to obtain an identity recognition result of the individual.

[0016] Compared with the prior art, the technical scheme of the present application has the following beneficial effects:

[0017] Based on the extraction of the spatial variability feature curve and the temporal variability feature curve of the electrocardiosignal, the present application models and approximates unknown nonlinear dynamic information and fast / slow change information under time-varying data, so that the first intrinsic dynamic feature extracted can better reflect the essential dynamic characteristics, thereby improving the recognition accuracy and success rate. In addition, according to the first recognition error, the present application uses an adaptive preset classifier scheme strategy, which can fully reflect the nonlinear characteristics of the electrocardiosignal and is also beneficial to comparing electrocardiosignals under different conditions, and has the advantages of low feature dimension, small computing resource demand, and fast computing speed. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The flowchart of the electrocardiosignal classification and recognition method of the present application embodiment 1 is shown.

[0019] Figure 2 The specified distance measurement under the Euclidean distance of the present application embodiment 1 is shown.

[0020] Figure 3 The specified distance measurement under the Manhattan distance of the present application embodiment 1 is shown.

[0021] Figure 4 The electrocardiosignal schematic diagram of the present application embodiment 2 is shown.

[0022] Figure 5 The spatial variability parameter schematic diagram of the present application embodiment 2 is shown.

[0023] Figure 6 The temporal variability parameter schematic diagram of the present application embodiment 2 is shown.

[0024] Figure 7Fig. 2 is a schematic diagram of a feature curve based on spatial variability of R wave for Embodiment 2 of the present application;

[0025] Figure 8 Fig. 4 is a schematic diagram of a feature curve based on spatial variability of T wave for Embodiment 2 of the present application;

[0026] Figure 9 Fig. 6 is a schematic diagram of a feature curve based on temporal variability of R wave for Embodiment 2 of the present application;

[0027] Figure 10 Fig. 8 is a schematic diagram of a feature curve based on temporal variability of T wave for Embodiment 2 of the present application;

[0028] Figure 11 Fig. 10 is a schematic diagram of a spatial running curve for the spatial variability feature curve for Embodiment 2 of the present application;

[0029] Figure 12 Fig. 12 is a schematic diagram of a spatial running curve for the temporal variability feature curve for Embodiment 2 of the present application. DETAILED DESCRIPTION

[0030] The terms "first", "second", and the like in the description and in the claims of the present application and in the above drawings are used for distinguishing between similar objects and not necessarily for describing a specific sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the present application are capable of functioning in other sequences, unless explicitly stated otherwise. Moreover, the terms "comprise", "have" and any variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises, has or includes a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0031] The accompanying drawings are only intended to be illustrative and should not be considered as limiting the present patent;

[0032] In order to better illustrate the present embodiments, some components in the accompanying drawings can be omitted, enlarged or reduced, and do not represent the actual size of the product;

[0033] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings can be omitted.

[0034] The technical solutions of the present application will be further described below in conjunction with the accompanying drawings and embodiments.

[0035] Embodiment 1

[0036] The present embodiment provides a method for classifying and recognizing electrocardiosignal based on spatio-temporal feature fusion, referring to Figure 1 , comprising:

[0037] acquiring an individual electrocardio signal, extracting a spatial variability feature curve and a temporal variability feature curve of the electrocardio signal;

[0038] constructing a three-dimensional state space, extracting an electrocardio spatiotemporal feature based on the spatial variability feature curve and the temporal variability feature curve of the electrocardio signal;

[0039] extracting a first intrinsic dynamics feature of the electrocardio signal based on the electrocardio spatiotemporal feature;

[0040] obtaining a first recognition error based on a difference between the first intrinsic dynamics feature and a second intrinsic dynamics feature stored in a preset electrocardio mode library; wherein the preset electrocardio mode library is constructed according to the electrodynamic signals of different individual identities under different preset conditions;

[0041] recognizing the electrocardio spatiotemporal feature to obtain an identity recognition result by using an adaptive preset classifier scheme according to the first recognition error.

[0042] In some examples, the electrocardio signal is acquired by a wearable device.

[0043] In some examples, the identity recognition result is a personal identity of a user;

[0044] In some examples, the identity recognition result is a new user or an old user;

[0045] In other examples, the identity recognition result is a risk level of a user, such as a low-risk population or a high-risk population.

[0046] In a preferred embodiment, the acquiring an individual electrocardio signal comprises:

[0047] acquiring the electrocardio signal of the individual at rest by using a twelve-lead electrocardio system;

[0048] performing filtering processing on the electrocardio signal to extract R waves and T waves in a cardiac cycle, and obtaining wave peak points of the R waves and the T waves in a plurality of cardiac cycles;

[0049] traversing the wave peak points of the R waves and the T waves, and for any adjacent R wave peak points and adjacent T wave peak points, extracting the spatial variability feature curve and the temporal variability feature curve of the electrocardio signal according to amplitude difference and time span difference between the wave peak points.

[0050] It can be understood that the electrocardio signal can be collected when the individual is at rest in a lying position, a standing position, or a sitting position, which can be determined by a person skilled in the art according to actual conditions.

[0051] In some examples, the filtering processing comprises average filtering and / or median filtering.

[0052] In an optional embodiment, the extracting the spatial variability characteristic curve and the time variability characteristic curve of the electrocardiosignal according to the amplitude difference value between adjacent peak points and the time span difference value comprises:

[0053] The amplitude difference value between adjacent peak points and the time span difference value are calculated, and the expression is:

[0054] Sr i = abs(y(r i )-y(r i+1 ))

[0055] St i = abs(y(t i )-y(t i+1 ))

[0056] Tr i = abs(x(r i )-x(r i+1 ))

[0057] Tt i = abs(x(t i )-x(t i+1 ))

[0058] In the formula, r i represents the i-th R-wave peak point; t i represents the i-th T-wave peak point; Sr i represents the amplitude difference value between adjacent two R-wave peak points in the spatial coordinate; St i represents the amplitude difference value between adjacent two T-wave peak points in the spatial coordinate; Tr i represents the time span difference value between adjacent two R-wave peak points in the time coordinate; Tt i represents the time span difference value between adjacent two T-wave peak points in the time coordinate; x(·) represents the horizontal coordinate value of the data point in the electrocardiosignal in the two-dimensional coordinate system; y(·) represents the vertical coordinate value of the data point in the electrocardiosignal in the two-dimensional coordinate system; abs(·) represents the absolute value operation;

[0059] The amplitude difference values Sr i and St i , and the time span difference values Tr i and Tt i are taken as the spatial variability parameters and the time variability parameters, and the characteristic curves of the spatial variability parameters and the time variability parameters changing with time, i.e., the spatial variability characteristic curve and the time variability characteristic curve, are constructed.

[0060] It can be understood that Sr i , St i , Tri , Tt i is a time interval of the electrocardiosignal, and the two respectively depict the variation of the electrocardiosignal over time from the spatial angle and the time angle.

[0061] In a preferred embodiment, the three-dimensional state space is constructed, the spatial variation degree feature curve and the time variation degree feature curve of the electrocardiosignal are extracted, including:

[0062] The three-dimensional state space is constructed, and the spatial variation degree feature curve and the time variation degree feature curve of the electrocardiosignal are reconstructed in phase space to obtain a spatial running curve about the spatial variation degree feature curve and a spatial running curve about the time variation degree feature curve, respectively.

[0063] The distance metric between all data points in the spatial running curve and the starting point of the data is calculated; wherein the distance metric includes one or more of the Euclidean distance, the cosine similarity, the Manhattan distance and the Minkowski distance.

[0064] The distance metric is taken as the electrocardiosignal electrocardiospatial-temporal feature.

[0065] It should be noted that in this preferred embodiment, through the phase space reconstruction operation, the spatial variation characteristics and the time variation characteristics of the electrocardiosignal are saved in the response space running curve, which can be drawn into a three-dimensional ring in a three-dimensional coordinate system, and the distance metric of the data point to the starting point of the data in the three-dimensional state space is taken as the representation form of the electrocardiospatial-temporal feature.

[0066] In some examples, for any kind of variation degree data sequence, that is, Sr i , Sr i , Tr i or Tt i , expressed as V={v1, v2, v3…, v k}, k represents the number of data points in each data sequence, which is reconstructed into {v j , v j+τ , …, v j+(d-1)τ} by phase space reconstruction; wherein j represents the embedding dimension required in the phase space reconstruction process, and τ represents the delay time.

[0067] In an optional embodiment, the electrocardiospatial-temporal feature is identified by using an adaptive preset classifier scheme according to the first identification error to obtain an identity recognition result, including:

[0068] It is judged whether the first identification error exceeds a preset value:

[0069] If yes, the selected preset classifier scheme includes: selecting a specified distance metric in the electrocardiogram spatiotemporal feature as a recognition feature, constructing a dynamic estimator according to a corresponding second intrinsic kinetic feature; subtracting the recognition feature from the dynamic estimator to obtain a second recognition error, taking the average L1 norm of the second recognition error as a third recognition error, and selecting an individual identity represented by the dynamic estimator corresponding to the smallest third recognition error as an identity recognition result;

[0070] Otherwise, the selected preset classifier scheme includes: selecting all distance metrics in the electrocardiogram spatiotemporal feature as recognition features to be respectively input into a one-dimensional convolutional neural network to obtain different deep feature matrices; concatenating all deep feature matrices to obtain a unified feature matrix, and using a specified classifier to perform recognition classification on the unified feature matrix to obtain an identity recognition result.

[0071] It should be noted that when the first recognition error does not exceed the preset value, the specified distance metric can be any one or more of Euclidean distance, cosine similarity, Manhattan distance, and Minkowski distance, which can be set by a person skilled in the art according to actual conditions. The dynamic estimator is constructed according to the corresponding second intrinsic kinetic feature, that is, if the Euclidean distance is selected as the specified distance metric, the subsequent dynamic estimator is also constructed according to the second intrinsic kinetic feature corresponding to the Euclidean distance.

[0072] In some examples, the specified distance metric is the Euclidean distance, see Figure 2 ;

[0073] In some examples, the specified distance metric is the Manhattan distance, see Figure 3 ;

[0074] In other examples, the specified distance metric includes the Euclidean distance and the Manhattan distance.

[0075] In some examples, the first recognition error is an absolute value, and feature extraction schemes A1, A2, A3, and A4 are defined according to the four different distance metrics of the Euclidean distance, the cosine similarity, the Manhattan distance, and the Minkowski distance, respectively.

[0076] The preset value is set to 0.5, and for the first recognition error not exceeding 0.5, that is, the first recognition error is in the interval [0, 0.5], the feature scheme S1 = {A1} is called, and the classifier scheme C1 is used:

[0077] A set of dynamic estimators is constructed according to A1, and the electrocardiogram spatiotemporal feature to be recognized is subtracted from the set of dynamic estimators to obtain a second recognition error.

[0078] The average L1 norm of the second identification error is taken as a third identification error, and based on the principle of minimum identification error, the individual identity represented by the dynamic estimator corresponding to the minimum third identification error is selected as the final identity recognition result;

[0079] For the first identification error exceeding 0.5, i.e. the first identification error is in the interval (0.5, +∞), a feature scheme S2 = {A1, A2, A3, A4} is called, and a classifier scheme C2 is adopted:

[0080] The four types of distance metrics corresponding to A1, A2, A3, and A4 are input into a one-dimensional convolutional neural network to obtain four different deep feature matrices, and a unified feature matrix is formed by matrix concatenation.

[0081] One or more specified classifiers are used to identify and classify the unified feature matrix to obtain an identity recognition result, thereby realizing the classification identification task.

[0082] In some examples, when the first identification error exceeds the preset value, the specified classifier can be a support vector machine (SVM), a K-nearest neighbor classifier (KNN), a decision tree, a naive Bayes classifier, or a variant thereof.

[0083] Further, the state expression of the dynamic estimator is:

[0084]

[0085] In the formula, represents the state of the dynamic estimator; B represents the parameter of the dynamic estimator; x represents the feature for identification; represents the first intrinsic kinetic feature; S(x) represents a preset radial basis function;

[0086] The expression of the second identification error is:

[0087]

[0088] In the formula, represents the second identification error; represents the second intrinsic kinetic feature; T represents a matrix transposition operation.

[0089] In a preferred embodiment, the first intrinsic kinetic feature of the electrocardiogram signal is extracted based on the electrocardiogram spatiotemporal feature, specifically:

[0090] A preset electrocardiogram dynamics network model is adopted The electrocardiogram spatiotemporal feature is approximated and modeled until the electrocardiogram dynamics network model converges, and the corresponding weight matrix of the converged electrocardiogram dynamics network model is taken as the first intrinsic kinetic feature;

[0091] wherein x = [x1,..., x n ] T ∈ R n represents electrocardiogram spatiotemporal feature data, n is the number of input data samples; p represents a constant parameter; F(x; p) = [f1(x; p),..., f n (x; p)] T represents a smooth and unknown electrocardiogram dynamics term; v(x; p) = [v1(x; p),..., v n (x; p)] T represents an introduced modeling uncertainty term.

[0092] It should be noted that the electrocardiogram dynamics term represents the fast or slow variation of the electrocardiogram signal over time, which is an important embodiment of the dynamic variation characteristics of the electrocardiogram signal. After the convergence of the electrocardiogram dynamics network model, the weights will converge to the optimal constant value, and all the weights are taken as the first intrinsic dynamics matrix in the form of a weight matrix.

[0093] In an optional embodiment, the first recognition error is obtained based on the difference between the first intrinsic dynamics feature and a second intrinsic dynamics feature stored in a preset electrocardiogram mode library, and the first recognition error comprises:

[0094] obtaining a preset electrocardiogram mode library; the preset electrocardiogram mode library stores at least one second intrinsic dynamics feature;

[0095] calculating the matrix difference between the first intrinsic dynamics feature and the second intrinsic dynamics feature to obtain at least one difference matrix;

[0096] calculating the average value of the elements in each difference matrix and performing an absolute value operation, taking the absolute value of each average value as a recognition error measurement value, and extracting the smallest recognition error measurement value ε as the first recognition error.

[0097] It should be noted that the calculation of the matrix difference between the first intrinsic dynamics feature and the second intrinsic dynamics feature is to subtract the element values at corresponding positions between the first intrinsic dynamics feature and the second intrinsic dynamics feature to construct a difference matrix.

[0098] In a preferred embodiment, the construction process of the preset electrocardiogram mode library comprises:

[0099] constructing an initial electrocardiogram mode library;

[0100] obtaining the heart signal of at least one individual under at least one preset condition, and extracting the spatial variability feature curve and the temporal variability feature curve of the electrocardiogram signal;

[0101] constructing a three-dimensional state space, and extracting the electrocardiogram spatiotemporal feature based on the spatial variability feature curve and the temporal variability feature curve of the electrocardiogram signal.

[0102] Based on the electrocardiospatial-temporal characteristics, the second intrinsic dynamic characteristics of the electrocardio signal are extracted and stored in the electrocardio mode library, and are labeled according to the individual identity.

[0103] It should be noted that the preset condition can be the period before and after the individual exercise, the period during which the individual is healthy or ill, the daytime or the night.

[0104] Embodiment 2

[0105] In this embodiment, the method described in embodiment 1 is simulated by using the PTB electrocardio database provided by the German National Metrology Institute Physikalisch-Technische Bundesanstalt (PTB). The recorded electrocardio data contains electrocardio records from 290 subjects, each record contains regular 12-lead (i, ii, iii, avr, avl, avf, v1, v2, v3, v4, v5, v6) data, as shown in Figure 4 .

[0106] The sampled continuous electrocardio signal time sequence is regarded as a data queue with a fixed length of N. The R wave and T wave in each cardiac cycle are extracted from the waveforms after the average filtering and median filtering processing, so as to obtain all the cardiac cycle R wave and T wave peak points.

[0107] Traverse all the electrocardio signals of the cardiac cycle: extract the i-th R wave peak point and record it as r i , extract the i-th T wave peak point and record it as t i ; calculate the amplitude difference and time span difference between adjacent R and T wave peak points for the electrocardio data sequence, and take them as spatial variability parameters and time variability parameters, as shown in Figure 5 , Figure 6 .

[0108] According to the spatial variability parameters and the time variability parameters, the characteristic curves of the spatial variability parameters and the time variability parameters changing with time are constructed, as shown in Figures 7-10 .

[0109] A three-dimensional state space is constructed, and the spatial variability characteristic curve and the time variability characteristic curve of the electrocardio signal are reconstructed in phase space, to obtain the spatial running curve about the spatial variability characteristic curve and the spatial running curve about the time variability characteristic curve, as shown in Figure 11 , Figure 12 ; the distance metric between all data points in the spatial running curve and the starting point of the data is calculated, and the distance metric is taken as the electrocardiospatial-temporal characteristics of the electrocardio signal.

[0110] a first recognition error is obtained based on a difference between the first intrinsic dynamics feature and a second intrinsic dynamics feature stored in a preset electrocardiogram mode library. The first intrinsic dynamics feature is obtained based on a training set of a PTB electrocardiogram database, and the second intrinsic dynamics feature is obtained based on a test set of the PTB electrocardiogram database and stored in the preset electrocardiogram mode library.

[0111] According to the first recognition error, the electrocardiogram space-time feature is recognized by using an adaptive preset classifier scheme to obtain an identity recognition result.

[0112] According to the above implementation process, in the recognition experiment in the PTB database, the recognition efficiency of the application can reach 95.7% at most.

[0113] Embodiment 3

[0114] The embodiment provides an electrocardiogram signal classification and recognition system based on space-time feature fusion. The method in Embodiment 1 is applied, and the system comprises the following steps:

[0115] The acquisition module is configured to acquire an electrocardiogram signal of an individual.

[0116] The feature extraction module is configured to extract a spatial variability feature curve and a temporal variability feature curve of the electrocardiogram signal, extract an electrocardiogram space-time feature based on the spatial variability feature curve and the temporal variability feature curve of the electrocardiogram signal, and extract a first intrinsic dynamics feature of the electrocardiogram signal based on the electrocardiogram space-time feature.

[0117] The identity recognition module is configured to carry a preset electrocardiogram mode library, the preset electrocardiogram mode library stores a second intrinsic dynamics feature, calculate a difference between the first intrinsic dynamics feature and the second intrinsic dynamics feature to obtain a first recognition error, and recognize the electrocardiogram space-time feature by using an adaptive preset classifier scheme according to the first recognition error to obtain an identity recognition result of the individual.

[0118] It can be understood that the system of the embodiment corresponds to the method in Embodiment 1, and the optional items in Embodiment 1 are also applicable to the embodiment, and thus are not described again here.

[0119] Embodiment 4

[0120] The embodiment provides a computer-readable storage medium, and the storage medium stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the method in Embodiment 1.

[0121] Exemplarily, the storage medium includes, but is not limited to, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.

[0122] Exemplarily, the instructions, programs, code sets or instruction sets can be implemented in programming languages such as Java, Python, C++, R, or Golang.

[0123] Exemplarily, the processor includes, but is not limited to, a smart phone, a personal computer, a server, a network device, etc., and is used to execute all or part of the steps of the electrocardiosignal classification and identification method described in Embodiment 1.

[0124] The same or similar reference signs correspond to the same or similar components;

[0125] The terms describing the positional relationship in the drawings are only used for exemplary illustration, and cannot be understood as a limitation on the patent;

[0126] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not a limitation on the embodiments of the present application. Based on the above description, other different forms of changes or variations can be made by those skilled in the art, each function module or unit can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. It is not necessary and impossible to enumerate all the embodiments. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A method for classifying and recognizing electrocardiogram signals based on spatiotemporal feature fusion, characterized in that, include: Acquire individual electrocardiogram (ECG) signals, and extract spatial variability characteristic curves and temporal variability characteristic curves of the ECG signals based on the amplitude difference and time span difference between the peaks of the ECG signals; A three-dimensional state space is constructed, and spatiotemporal features of electrocardiogram (ECG) signals are extracted based on the spatial and temporal variability characteristic curves. Based on the spatiotemporal characteristics of electrocardiogram (ECG), the first intrinsic dynamic feature of ECG signal is extracted; The first identification error is obtained based on the difference between the first intrinsic dynamic feature and the second intrinsic dynamic feature stored in the preset ECG pattern library; wherein, the preset ECG pattern library is constructed according to the cardiac signals of individuals with different identities under different preset conditions; Based on the first recognition error, an adapted preset classifier scheme is used to identify the spatiotemporal features of the electrocardiogram to obtain the identity recognition result; The extraction of spatial variability feature curves and temporal variability feature curves of electrocardiogram signals includes: The expression for calculating the amplitude difference and time span difference between adjacent wave crests is as follows: Sr i =abs(y(r i )-y(r i+1 )) St i =abs(y(t i )-y(t i+1 )) Tr i =abs(x(r i )-x(r i+1 )) Tt i =abs(x(t i )-x(t i+1 )) In the formula, r i t represents the i-th R-wave peak; i Sr represents the i-th T-wave peak; i St represents the amplitude difference between two adjacent R-wave peaks in spatial coordinates; i This represents the amplitude difference between two adjacent T-wave peaks in spatial coordinates; Tr i Tt represents the time span difference between two adjacent R-wave peaks on the time coordinate; i This represents the time span difference between two adjacent T-wave peaks on the time coordinate; x(·) represents the abscissa value of the data point in the ECG signal in the two-dimensional coordinate system; y(·) represents the ordinate value of the data point in the ECG signal in the two-dimensional coordinate system; abs(·) represents the absolute value operation; The amplitude difference Sr i and St i , and the difference in time span Tr i and Tt i As spatial variability parameters and temporal variability parameters, characteristic curves of the spatial variability parameters and temporal variability parameters as a function of time are constructed, namely, spatial variability characteristic curves and temporal variability characteristic curves. And, the step of using an adapted preset classifier scheme to identify the spatiotemporal features of the electrocardiogram based on the first identification error to obtain the identity recognition result includes: Determine whether the first recognition error does not exceed the preset value: If so, the selected preset classifier scheme includes: selecting a specified distance metric in the spatiotemporal features of electrocardiogram as the identification feature, constructing a dynamic estimator based on the corresponding second intrinsic dynamic feature; subtracting the identification feature from the dynamic estimator to obtain the second identification error, using the average L1 norm of the second identification error as the third identification error, and selecting the individual identity represented by the dynamic estimator corresponding to the smallest third identification error as the identity recognition result. Otherwise, the selected preset classifier scheme includes: selecting all distance metrics in the ECG spatiotemporal features as recognition features and inputting them into a one-dimensional convolutional neural network to obtain different depth feature matrices; concatenating all depth feature matrices to obtain a unified feature matrix; and using a specified classifier to perform recognition and classification based on the unified feature matrix to obtain the identity recognition result. And, the first identification error is obtained based on the difference between the first intrinsic dynamic feature and the second intrinsic dynamic feature stored in the preset ECG pattern library, including: Obtain a preset ECG pattern library; the preset ECG pattern library stores at least one second intrinsic dynamic feature; Calculate the matrix difference between the first intrinsic dynamic feature and the second intrinsic dynamic feature to obtain at least one difference matrix; Calculate the average value of each element in the difference matrix and take its absolute value. Use the absolute value of each average value as the recognition error metric and extract the smallest recognition error metric ε as the first recognition error.

2. The electrocardiogram signal classification and recognition method based on spatiotemporal feature fusion according to claim 1, characterized in that, The acquisition of individual electrocardiogram signals includes: A twelve-lead electrocardiogram system was used to collect individual resting electrocardiogram signals. The electrocardiogram signal is filtered to extract the R wave and T wave in the cardiac cycle, and the peak points of the R wave and T wave in several cardiac cycles are obtained. Traverse the peaks of the R and T waves. For any adjacent R wave peak and adjacent T wave peak, extract the spatial variability characteristic curve and temporal variability characteristic curve of the electrocardiogram signal based on the amplitude difference and time span difference between the peaks.

3. The electrocardiogram signal classification and recognition method based on spatiotemporal feature fusion according to claim 1, characterized in that, The construction of the three-dimensional state space, based on the spatial and temporal variability characteristic curves of the electrocardiogram (ECG) signal, extracts spatiotemporal features of the ECG, including: A three-dimensional state space is constructed, and the spatial variability characteristic curves and temporal variability characteristic curves of the electrocardiogram signal are reconstructed in phase space to obtain spatial operation curves with respect to the spatial variability characteristic curves and temporal variability characteristic curves, respectively. Calculate the distance metric between all data points within the spatial running curve and the starting point of the data; wherein the distance metric includes one or more of Euclidean distance, cosine similarity, Manhattan distance, and Minkowski distance; Distance metric is used as the spatiotemporal feature of electrocardiogram (ECG) signals.

4. The electrocardiogram signal classification and recognition method based on spatiotemporal feature fusion according to claim 3, characterized in that, The state expression of the dynamic estimator is: In the formula, Indicates the state of the dynamic estimator; k represents the index number; B represents the parameters of the dynamic estimator; x represents the features used for identification; This represents the first intrinsic dynamic characteristic; S(x) represents the preset radial basis function; The expression for the second identification error is: In the formula, This indicates the second identification error; This indicates the second intrinsic dynamic characteristic; T indicates that the matrix transpose operation is performed.

5. The electrocardiogram signal classification and recognition method based on spatiotemporal feature fusion according to claim 1, characterized in that, Based on the spatiotemporal characteristics of electrocardiogram (ECG), the first intrinsic dynamic feature of the ECG signal is extracted. Specifically: Using a pre-defined electrocardiographic dynamics network model The spatiotemporal features of electrocardiogram are approximated and modeled until the electrocardiogram dynamic network model converges. The corresponding weight matrix of the converged electrocardiogram dynamic network model is used as the first intrinsic dynamic feature. Where x = [x1, ..., x n ] T ∈R n This represents the spatiotemporal feature data of electrocardiogram, where n is the number of input data samples; p represents a constant parameter; F(x; p) = [f1(x; p), ..., f n (x;p)] T Represents a smooth and unknown electrocardiographic term; v(x; p) = [v1(x; p), ..., v n (x;p)] T This indicates the introduced modeling uncertainty.

6. A method for classifying and recognizing electrocardiogram signals based on spatiotemporal feature fusion according to any one of claims 1-5, characterized in that, The construction process of the preset ECG pattern library includes: Build an initial ECG pattern library; Acquire cardiac signals of at least one individual under at least one preset condition, and extract the spatial variability characteristic curve and temporal variability characteristic curve of the electrocardiogram signal; A three-dimensional state space is constructed, and spatiotemporal features of electrocardiogram (ECG) signals are extracted based on the spatial and temporal variability characteristic curves. Based on the spatiotemporal characteristics of electrocardiogram (ECG), the second intrinsic dynamic features of ECG signals are extracted and stored in the ECG pattern library, and labeled according to individual identity.

7. A classification and recognition system for electrocardiogram signals based on spatiotemporal feature fusion, using the method described in any one of claims 1-6, characterized in that, include: The acquisition module is used to acquire an individual's electrocardiogram (ECG) signals. The feature extraction module is used to extract the spatial variability feature curve and the temporal variability feature curve of the electrocardiogram (ECG) signal; it is also used to extract the spatiotemporal features of the ECG signal based on the spatial variability feature curve and the temporal variability feature curve of the ECG signal; and it is also used to extract the first intrinsic dynamic feature of the ECG signal based on the spatiotemporal features of the ECG signal. The identity recognition module is used to carry a preset ECG pattern library, which stores a second intrinsic dynamic feature; it is also used to calculate the difference between the first intrinsic dynamic feature and the second intrinsic dynamic feature to obtain a first recognition error; and it is also used to identify the spatiotemporal features of ECG using an adapted preset classifier scheme based on the first recognition error to obtain the individual's identity recognition result.

Images