Electrocardio identity recognition method and system based on adaptive weighted feature fusion
Through the ECG identity recognition method of adaptive weighted feature fusion, combined with local and global feature extraction modules, the problems of poor long-range feature extraction ability and high computational complexity in ECG signal recognition are solved, and efficient and accurate ECG identity recognition is achieved.
Patent Information
- Application Number
- CN202510547980.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-28
- Publication Date
- 2025-09-16
AI Technical Summary
Existing ECG signal recognition technology has problems such as poor long-range feature extraction capability, inability to conduct parallel training, and excessive computational complexity, which limits its application in real-time biometric recognition tasks.
An ECG identity recognition method based on adaptive weighted feature fusion is adopted. Through the hybrid architecture of local feature extraction module and global feature extraction module, combined with the adaptive weight feature fusion module, the local morphological features and global temporal dependencies of ECG signals are jointly modeled, and regularization constraints are performed through adaptive weight balance loss.
It significantly improves the accuracy and robustness of ECG signal recognition, reduces computational complexity, and improves the precision and efficiency of identity recognition, making it suitable for real-time biometric recognition tasks.
Smart Images
Figure CN120653943A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of identity recognition, and in particular relates to an electrocardiogram identity recognition method and system based on adaptive weighted feature fusion. Background Art
[0002] With the accelerating global digitalization process, the value and importance of information data are becoming increasingly prominent, and the security of personal information is becoming increasingly critical. Against this backdrop, traditional cryptographic-based identity authentication systems (such as passwords, PIN codes, and gesture verification) are no longer able to meet the stringent requirements of modern digital society for information protection and security due to inherent security flaws (e.g., brute force attacks or physical compromise). Consequently, biometric recognition technologies (such as fingerprints, faces, and irises), due to their ability to effectively reflect an individual's unique physiological characteristics, have been widely adopted and accepted in identity recognition systems. However, with the rapid development of generative adversarial networks and deepfake technologies, these static biometrics face an increasingly severe risk of counterfeiting. Against this backdrop, biometric recognition technologies based on electrocardiograms (ECGs) have gradually become a research focus in the field of identity recognition due to their unique advantages.
[0003] ECG signals are essentially a surface representation of the electrophysiological activity of cardiomyocytes. Their waveform characteristics (P wave, QRS complex, T wave, etc.) reflect individual differences in the heart's internal structure. Unlike other biometrics, ECG signal generation relies on the physiological continuity of cardiac electrical activity, making them remarkably live and inherently difficult to imitate or replicate. More importantly, ECG signals are temporally stable and permanent, maintaining consistency over extended periods, making them significantly more secure than traditional biometrics. These characteristics make ECG signals a more reliable biometric for identity verification.
[0004] Early studies primarily located key ECG waveform reference points (P wave, QRS complex, T wave) to extract time-domain features (such as the RR interval, QT interval, ST-segment slope) and morphological parameters (amplitude, integrated area, and waveform curvature), or combined them with traditional machine learning algorithms for classification. With the rapid development of deep learning technology, researchers have gradually shifted to an end-to-end feature learning paradigm: Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and various network variants (CNN-LSTM, HLSTM, BiLSTM, CNN-BiLSTM, RNN-Bi-LSTM, etc.) have been designed and applied to automatic feature extraction from ECG signals. While these methods have achieved promising performance, they still face several challenges: the local receptive field of CNNs makes it difficult to effectively capture long-range dependencies; LSTM models, due to their temporal recursion, cannot process sequential data in parallel, and the long-term dependency decay problem persists. Although Transformer achieves global context modeling through the self-attention mechanism, its quadratic computational complexity faces a serious efficiency bottleneck when processing long sequences of ECG signals, limiting its application in real-time biometric recognition tasks. Summary of the Invention
[0005] The purpose of the present invention is to overcome the problems of poor long-range feature extraction capability, inability to conduct parallel training and excessive computational complexity in the prior art, and propose an ECG identity recognition method and system based on adaptive weighted feature fusion.
[0006] In a first aspect, the present invention provides an electrocardiogram (ECG) identity recognition method based on adaptive weighted feature fusion, the method comprising: Collecting the subject's ECG signals and constructing a data set based on the collected ECG signals; constructing an ECG identity recognition model, and using the data set to train the constructed ECG identity recognition model, and achieving identity recognition through the trained ECG identity recognition model; The ECG identity recognition model includes a local feature extraction module, a global feature extraction module, an adaptive weight feature fusion module and a classifier; the local feature extraction module and the global feature extraction module respectively extract local features and global features of the ECG signal, and input them into the adaptive weight feature fusion module for fusion to obtain fusion features; the fusion features are input into the classifier for classification to obtain classification results; The global feature extraction module includes one or more global feature extraction layers connected in series and a global average pooling layer; the input features of the global feature extraction module are processed by one or more global feature extraction layers connected in series, and the processing results are input to the global average pooling layer to obtain the output of the global feature extraction module; The global feature extraction layer includes a first linear layer, a convolutional layer, a state space model block and a second linear layer; after the input features of the global feature extraction module are processed by the first linear layer and the convolutional layer in sequence, the input features are input into the state space model block; the output features of the state space model block are fused with the output features of the first linear layer after being processed by the activation function, and the fusion results are processed by the second linear layer to obtain the output features of the global feature extraction layer.
[0007] Preferably, the local feature extraction module includes a local feature extraction layer, a first residual block, a second residual block, a third residual block and a global average pooling layer connected in sequence; the number of channels of the convolution layer in adjacent residual blocks increases sequentially, and the step size decreases sequentially.
[0008] Preferably, the local feature extraction layer includes a convolutional layer, a batch normalization layer, and an activation function connected in sequence; the first residual block, the second residual block, and the third residual block each include multiple residual layers connected in series; the input feature of each residual layer is used as the first feature, and the output feature of the residual layer is processed in sequence using batch normalization and activation function to obtain the second feature; the feature obtained by fusion of the first feature and the second feature is used as the input of the next residual layer; If the dimensions of the first feature and the second feature do not match, the first feature is convolved; the fusion result of the first feature and the second feature after the convolution process is used as the input of the next residual layer.
[0009] As an advantage, the adaptive weight feature fusion module defines the learnable weight coefficients of the local feature extraction module and the global feature extraction module respectively. and , and the normalized learnable weight coefficient and learnable weight coefficients The output features of the local feature extraction module and the global feature extraction module are fused; the fusion results are processed by the linear layer and activation function in sequence to obtain the fusion features output by the adaptive weight feature fusion module.
[0010] Preferably, during the training process, cross entropy loss and adaptive weight balance loss are used to construct the total loss function of the ECG identity recognition model.
[0011] As a preference, the above-mentioned adaptive weight balancing loss The construction method is as follows: .
[0012] Preferably, before constructing the data set, the collected ECG signals are preprocessed; the preprocessing includes noise removal, R-peak detection and heart beat segmentation.
[0013] In the second aspect, the present invention provides an ECG identity recognition system based on adaptive weighted feature fusion, which is used to execute the above-mentioned ECG identity recognition method; the ECG identity recognition system includes a signal acquisition module, a preprocessing module and an ECG identity recognition module; the signal acquisition module is used to collect the ECG signal of the subject; the preprocessing module is used to preprocess the collected ECG signal; and the ECG identity recognition module is used to perform identity recognition on the preprocessed ECG signal.
[0014] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores the computer program; and the processor executes the above-mentioned electrocardiogram identity recognition method.
[0015] In a fourth aspect, the present invention provides a readable storage medium storing a computer program; when the computer program is executed by a processor, it is used to implement the above-mentioned electrocardiogram identity recognition method.
[0016] The present invention has the following beneficial effects: 1. The present invention performs global feature extraction on ECG signals through a global feature extraction module, significantly enhancing the ability to process time series features and improving the accuracy of extracting identity-related features in ECG signals; at the same time, the present invention extracts deep-level temporal information by setting convolution kernels with different step lengths and numbers of channels in the local feature extraction module, thereby improving the feature expression ability of the ECG identity recognition model and effectively alleviating the gradient vanishing problem in the deep network; in addition, the present invention realizes the joint modeling of local morphological features and global temporal dependencies of ECG signals while reducing computational complexity and improving computational efficiency through a hybrid architecture of local feature extraction modules and global feature extraction modules, thereby improving the accuracy and robustness of identity recognition and having broad practical application value.
[0017] 2. The present invention introduces a learnable weight coefficient through an adaptive weight feature fusion module, and performs regularization constraints by constructing a weight balance loss, which effectively solves the modal deviation problem caused by static weights in traditional feature fusion methods and the weight polarization problem in small sample scenarios, and improves the accuracy of ECG identity recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. The drawings described below are some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.
[0019] Figure 1 It is the overall flow chart of the present invention.
[0020] Figure 2 Schematic diagram of the structure of the local feature extraction module in the present invention.
[0021] Figure 3 Schematic diagram of the structure of the state space block in the present invention. DETAILED DESCRIPTION
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0023] A method for ECG identity recognition based on adaptive weighted feature fusion uses an ECG identity recognition system comprising a signal acquisition module, a preprocessing module, and an ECG identity recognition module. The signal acquisition module is used to acquire the subject's ECG signal; the preprocessing module is used to preprocess the acquired ECG signal; and the ECG identity recognition module is used to perform identity recognition based on the preprocessed ECG signal.
[0024] like Figure 1 As shown, the ECG identity recognition method includes the following steps: Step 1: Collect the subject's ECG signal as the original ECG signal, preprocess the original ECG signal to obtain the preprocessed ECG signal, i.e., the double heartbeat signal, and construct a data set based on the preprocessed ECG signal; the specific process of data preprocessing is as follows: 1-1. Noise Cancellation Raw ECG signals are often subject to noise interference from multiple sources: 0.05-2 Hz baseline drift, caused by changes in electrode-skin interface impedance due to respiratory movement; 50 / 60 Hz power frequency interference, generated by electromagnetic coupling from the power system; and 5-500 Hz electromyographic artifacts, random high-frequency noise from skeletal muscle contraction. Therefore, a Butterworth filter with a sampling rate of 500 Hz and cutoff frequencies of 3 Hz and 45 Hz is used to process the raw ECG signals. This filter removes these interferences while retaining the main components of the ECG signal.
[0025] 1-2. R peak detection The Hamilton algorithm (Hamilton graph algorithm) is used on the ECG signal after noise removal to obtain the R peak position of each heart beat.
[0026] 1-3. Heart beat segmentation Based on the detected R-peak positions, heartbeat segmentation and normalization are further performed. A total of 880 sample points (corresponding to 1.76 seconds at 500Hz sampling) are captured, starting from 200 samples before the first R-peak and ending 240 samples after the second R-peak. This constitutes the dual-beat signal input, which serves as the normalized input for the subsequent dual encoder network. If the sampling rate of the dataset is higher than 500Hz, it is downsampled to ensure consistency during data processing and meet model input requirements.
[0027] Step 2: Build an ECG identity recognition model The ECG identity recognition model includes a local feature extraction module, a global feature extraction module, an adaptive weight feature fusion module, and a classifier. The local feature extraction module is used to extract local feature information of the ECG signal, which includes a local feature extraction layer, a first residual block, a second residual block, a third residual block, and a global average pooling layer connected in sequence. The local feature extraction layer is used to perform the preprocessed ECG signal (input signal) Perform time series downsampling operation and output primary features ;in, B is the batch size; L is the length of the ECG signal; D is the number of channel dimensions of the ECG signal; the local feature extraction layer includes a convolutional layer, a batch normalization layer, and a RELU activation function connected in sequence; the convolution kernel size of the convolution layer in the local feature extraction layer is 1×32, the number of channels is 16, and the stride is 2.
[0028] like Figure 2 As shown, the first residual block includes three stacked residual layers; the second residual block includes four stacked residual layers; and the third residual block includes three stacked residual layers. The residual layers are connected using the identity mapping optimization strategy; except for the first residual layer in the second residual block and the third residual block, the output features of each residual layer are processed by batch normalization and ReLU activation function in turn, and then fused with the input features of the residual layer, and the fusion results are used as the input of the next residual layer; the output features of the first residual layer in the second residual block and the third residual block are processed by batch normalization and ReLU activation function in turn, and the input features of the residual layer are processed by 1×1 convolution kernel, and the processed input features and output features are fused as the input features of the next residual layer; among them, the batch normalization layer performs zero mean unit variance standardization on the feature distribution, suppresses the internal covariance shift, accelerates the training process and improves the stability of the model, and the ReLU activation function introduces nonlinear expression ability to accelerate model convergence. Input features of each residual layer It is expressed as follows: (1) in, For the The input features of the residual layer; is the residual function, which represents the nonlinear transformation operation in the main path; This stands for projection mapping. When the input and output dimensions do not match, a 1×1 convolution kernel is used to adjust the input feature dimensions to ensure additivity of the residual connection. If the dimensions match, it degenerates into an identity mapping. This design ensures that gradients can be passed unimpeded through skip connections during backpropagation, effectively alleviating the gradient vanishing problem in deep networks.
[0029] Each residual layer consists of two convolutional layers connected in sequence. Deep temporal information is extracted by setting convolution kernels with different step sizes and numbers of channels in different residual blocks. The convolution kernel size of the convolution layer in the first residual block, the second residual block, and the third residual block are all 1×7, with 16, 32, and 64 channels, and step sizes of L / 2, L / 4, and L / 8, respectively. The last residual layer in the third residual block adds the input and output of this layer through a jump connection, and after being processed by the batch normalization layer and the RELU activation function, it is input into the global average pooling layer. The global average pooling layer is used to compress the extracted features along the temporal dimensionality reduction to reduce the size of the features and output the final local features. The local feature extraction module gradually extracts local features of the ECG signal through a series of convolutional layers, skip connections, and pooling operations. The design of each layer not only aims to improve the network's feature expression capabilities, but also ensures effective information transmission through skip connections, effectively alleviating the vanishing gradient problem in deep networks.
[0030] The global feature extraction module is used to perform global modeling of ECG signals. It consists of two global feature extraction layers and a global average pooling layer connected in sequence. The two global feature extraction layers have the same structure, including a first linear layer, a convolutional layer, a state space model (SSM) block, and a second linear layer. The first linear layer processes the input features. ( B is the batch size, L is the input sequence length, D For each time step t The hidden layer dimension is linearly projected to expand the input dimension to 2 ED ;in, E is the expansion factor. In this embodiment, E = 2. The output of the first linear layer is processed by a one-dimensional convolutional layer and a SiLU activation function, and then input into the state-space model block.
[0031] like Figure 3 As shown, given a one-dimensional input state space model block and N-dimensional hidden state , the system evolution process can be expressed as a linear ordinary differential equation: (2) (3) in, is the state transition matrix, is the input projection matrix, is the observation matrix, is the feedforward matrix. Can be regarded as a skip connection and does not affect the hidden state , so this term is omitted (let ). Subsequently, the continuous equation is discretized using the Zero-Order Hold (ZOH) method to obtain the recursive formula: (4) (5) Based on input , and calculated by discretization method 、 , the discrete parameters satisfy , ,in Represents the step size. By iteratively expanding formulas (4)-(5), the system output can be reconstructed into a convolution kernel The convolution operation with the input (Formulas (9)-(10)) is expressed as follows: (6) (7) (8) … (9) (10) in, Represents the convolution operation.
[0032] The output of the second linear layer is processed by the SiLU activation function and then combined with the output features of the state space model block The fusion is finally processed by the second linear layer to obtain the global features output by the global feature extraction layer .
[0033] The core idea of the adaptive weight feature fusion module is to achieve adaptive calibration of feature contribution through a dynamic weight allocation mechanism to achieve effective feature fusion, thereby coordinating the heterogeneous feature expressions of the two feature extraction modules. The adaptive weight feature fusion module effectively solves the modal bias problem caused by static weights in traditional feature fusion methods by introducing learnable weight coefficients and regularization constraints. For the local feature extraction module and the global feature extraction module, the corresponding learnable weight coefficients are defined respectively. and , in order to ensure the stability of the two weight coefficients, the learnable weight coefficients are and learnable weight coefficients Perform normalization: (11) (12) in, and are the learnable weight coefficients corresponding to the normalized local feature extraction module and the global feature extraction module respectively.
[0034] By normalizing the learnable weight coefficients, we ensure , thereby constraining the weight coefficients within the probability space and enhancing the training stability. Subsequently, the output features of the local feature extraction module and the global feature extraction module are dynamically weighted fused through the normalized weights, and the fusion results are It is expressed as follows: (13) in, and They are local features and global features respectively; Represents a splicing operation; Represents a channel-by-channel multiplication operation.
[0035] The fusion results After being processed by the linear layer and ReLU activation function in sequence, the spliced features are mapped to the target dimension through the fully connected layer, and the ReLU activation function is used to enhance the nonlinear expression ability to obtain the fusion features output by the adaptive weight feature fusion module. , which is expressed as: (14) in, represents the activation function; represents a linear layer.
[0036] The classifier is used to fusion features Mapping to the target category space, thus completing the classification and outputting the final classification result. The classifier sequentially uses a linear layer containing N classification heads and a Softmax function to output features. Processing to obtain the final prediction results , which is expressed as: (15) in, It is a linear layer that maps the feature dimension from D to the number of categories N; The function implements category probability normalization.
[0037] Step 3: Training the model The ECG identification model constructed in step 2 is trained using the dataset constructed in step 1. During the training process, a multi-objective joint optimization strategy is used to construct the total loss function. , which is expressed as follows: (16) in, is the cross entropy loss; Balancing loss for adaptive weights; A hyperparameter that controls the penalty strength of the adaptive weight balancing loss.
[0038] Cross Entropy Loss It is expressed as follows: (17) in, N is the number of samples in the dataset; is the true value.
[0039] Adaptive weight balancing loss To address the weight polarization phenomenon that may occur in small sample training scenarios (such as or ), which is expressed as follows: (18) This loss function minimizes the squared deviation between the weight coefficient and the equilibrium value (1), thus preventing the model from over-relying on a certain feature during a certain iteration, which may lead to unstable training process and the risk of overfitting. Compared with traditional feature fusion strategies (such as direct concatenation or average weighting), this module automatically learns through backpropagation and can adjust the contribution ratio of local and global features according to the characteristics of the input signal.
[0040] Adam optimizer is used, and the initial learning rate is set to 1e -3 , weight decay is set to 1e -4 , penalty intensity hyperparameter The learning rate is set to 0.5. To dynamically adjust the learning rate during training, the ReduceLROnPlateau learning rate scheduler is used. This scheduler monitors the loss on the dataset and automatically reduces the learning rate if the loss does not improve within 5 consecutive epochs. The total number of training rounds is set to 80 epochs, with a batch size of 32.
[0041] Step 4: Using the present invention, ECG identity recognition was performed on different datasets (ECG-ID, PTB, CYBHi, and ECG-Lab). Accuracy (Accuracy), precision (Precision), recall (Recall), and harmonic mean (F1-score) were used as performance evaluation parameters. Table 1 shows the performance of the proposed method, demonstrating good performance and effectiveness. CYBHi-L is a mid- to long-term CYBHi dataset.
[0042] Table SEQ Table\* ARABIC 1 Performance of the present invention on different data sets The present invention and existing ECG identity recognition methods (LSTM, CNN, LSTM, CNN-LSTM, Transformer) are used for ECG identity recognition respectively, and the parameter quantity (Param) and accuracy are introduced to evaluate the recognition results of different ECG identity recognition methods. The evaluation results are shown in Table 2.
[0043] Table 2 Comparison results between the present invention and other existing ECG identification methods As shown in Table 2, the proposed method achieves 100% recognition accuracy with only 1.2 times the number of parameters (1.24M) of the basic LSTM model (1.03M). Compared to the CNN-LSTM model (3.26M), which has 2.6 times the number of parameters, the accuracy is improved by 0.93%. Furthermore, the proposed method uses only 1 / 8 the number of parameters of the Transformer architecture (9.33M). This result demonstrates that the proposed method significantly reduces the computational overhead while maintaining high accuracy.
[0044] The above embodiments are not limitations of the present invention, and the present invention is not limited to the above embodiments. As long as the requirements of the present invention are met, they belong to the protection scope of the present invention.
Claims
1. An electrocardiogram (ECG) identity recognition method based on adaptive weighted feature fusion, characterized by: The method comprises: Collecting the subject's ECG signals and constructing a data set based on the collected ECG signals; constructing an ECG identity recognition model, and using the data set to train the constructed ECG identity recognition model, and achieving identity recognition through the trained ECG identity recognition model; The ECG identity recognition model includes a local feature extraction module, a global feature extraction module, an adaptive weight feature fusion module and a classifier; the local feature extraction module and the global feature extraction module respectively extract local features and global features of the ECG signal, and input them into the adaptive weight feature fusion module for fusion to obtain fusion features; the fusion features are input into the classifier for classification to obtain classification results; The global feature extraction module includes one or more global feature extraction layers connected in series and a global average pooling layer; the input features of the global feature extraction module are processed by one or more global feature extraction layers connected in series, and the processing results are input to the global average pooling layer to obtain the output of the global feature extraction module; The global feature extraction layer includes a first linear layer, a convolutional layer, a state space model block and a second linear layer; after the input features of the global feature extraction module are processed by the first linear layer and the convolutional layer in sequence, the input features are input into the state space model block; the output features of the state space model block are fused with the output features of the first linear layer after being processed by the activation function, and the fusion results are processed by the second linear layer to obtain the output features of the global feature extraction layer.
2. The ECG identity recognition method based on adaptive weighted feature fusion according to claim 1, characterized in that: The local feature extraction module includes a local feature extraction layer, a first residual block, a second residual block, a third residual block and a global average pooling layer connected in sequence; the number of channels of the convolution layer in adjacent residual blocks increases sequentially and the step size decreases sequentially.
3. The ECG identity recognition method based on adaptive weighted feature fusion according to claim 2, characterized in that: The local feature extraction layer includes a convolutional layer, a batch normalization layer, and an activation function connected in sequence; the first residual block, the second residual block, and the third residual block each include multiple residual layers connected in series; the input feature of each residual layer is used as the first feature, and the output feature of the residual layer is processed in sequence using batch normalization and an activation function to obtain a second feature; the feature obtained by fusing the first feature and the second feature is used as the input of the next residual layer; If the dimensions of the first feature and the second feature do not match, the first feature is convolved; the fusion result of the first feature and the second feature after the convolution process is used as the input of the next residual layer.
4. The ECG identity recognition method based on adaptive weighted feature fusion according to claim 1, characterized in that: The adaptive weight feature fusion module defines the learnable weight coefficients of the local feature extraction module and the global feature extraction module respectively. and , and the normalized learnable weight coefficient and learnable weight coefficients Fusing the output features of the local feature extraction module and the global feature extraction module; The fusion results are processed by the linear layer and activation function in sequence to obtain the fusion features output by the adaptive weight feature fusion module.
5. The ECG identity recognition method based on adaptive weighted feature fusion according to claim 1, characterized in that: During the training process, cross entropy loss and adaptive weight balance loss are used to construct the total loss function of the ECG identity recognition model.
6. The ECG identity recognition method based on adaptive weighted feature fusion according to claim 1, characterized in that: The adaptive weight balancing loss mentioned above The construction method is as follows: 。 7. The ECG identity recognition method based on adaptive weighted feature fusion according to claim 1, characterized in that: Before constructing the dataset, the collected ECG signals are preprocessed; the preprocessing includes noise removal, R-peak detection and heart beat segmentation.
8. An electrocardiogram identity recognition system based on adaptive weighted feature fusion, characterized by: Used to execute the ECG identity recognition method based on adaptive weighted feature fusion as described in claim 1; the ECG identity recognition system includes a signal acquisition module, a preprocessing module and an ECG identity recognition module; The signal acquisition module is used to collect the subject's electrocardiogram signal; The preprocessing module is used to preprocess the collected ECG signals; the ECG identity recognition module is used to perform identity recognition on the preprocessed ECG signals.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The memory stores a computer program; and the processor executes the electrocardiogram identity recognition method according to any one of claims 1 to 7.
10. A readable storage medium storing a computer program; characterized in that: When the computer program is executed by a processor, it is used to implement the electrocardiogram identity recognition method according to any one of claims 1 to 7.