Identity authentication method based on mixed physiological state electrocardiosignal

Abstract

The invention discloses an identity authentication method based on electrocardiosignals in a mixed physiological state, relates to the technical field of biological feature recognition, and aims to solve the problems that in the prior art, the research on identity recognition of electrocardiosignals mainly focuses on a resting state, and the research on identity recognition before and after movement is less. The method comprises the following steps: acquiring mixed electrocardiosignal data of a to-be-tested person; preprocessing the mixed electrocardiosignal data to obtain preprocessed mixed electrocardiosignal data; taking the preprocessed mixed electrocardiosignal data as input, and extracting network model output based on the electrocardiosignal to obtain an embedded feature vector; and calculating a cosine similarity value according to the embedded feature vector, and authenticating the identity of the person to be tested according to the cosine similarity value. The method can adapt to and process the electrocardiosignals in the motion state and the resting state at the same time, and the accuracy and robustness of identity authentication are improved.

Description

Technical Field

[0001] The invention relates to an identity authentication method based on mixed physiological state electrocardiogram signals, belonging to the technical field of biological feature recognition. Background Art

[0002] Biometrics is an important means of identity authentication in today's society. Mainstream biometrics include fingerprint, palm print, face, iris, etc. However, these traditional identification methods have certain limitations: fingerprint and palm print are easily affected by surface conditions and are at risk of being copied; face recognition accuracy will decrease under factors such as lighting changes and aging; although iris recognition is highly accurate, it requires professional data collection equipment and is relatively expensive.

[0003] As one of the physiological characteristics of the human body, ECG signals have unique advantages. Since the position, size and structure of each person's heart are different, ECG signals have significant individual differences. At the same time, ECG signals are bioelectric signals actively generated by the human body, have living characteristics, and are difficult to forge. However, the current identity authentication technology based on ECG signals still has technical bottlenecks. ECG signals are easily affected by external factors such as emotions and movement. In particular, ECG signals in motion change significantly compared to those in resting state, resulting in a decrease in recognition accuracy.

[0004] In recent years, with the development of deep learning technology, ECG signal identification methods based on deep neural networks have made significant progress. Existing methods mainly include feature extraction based on convolutional neural networks (CNNs) and time series modeling based on recurrent neural networks (RNNs). However, these methods still have the following problems when processing ECG signals in motion: first, the traditional single-scale convolution structure is difficult to capture the local detail features and long-range dependencies of ECG signals at the same time; second, there is a lack of effective feature selection mechanism, which cannot highlight the contribution of important feature channels; third, the training strategy does not fully consider the distribution differences of samples in motion and resting states, resulting in inconsistent performance of the model in different states.

[0005] At present, the research on ECG signal identification at home and abroad is mainly focused on the resting state, and there are relatively few studies on identification before and after exercise. Exercise can cause problems such as heart rate changes and ECG waveform distortion, which makes the identification algorithm developed based on the resting state perform poorly in the moving state. At the same time, there is a lack of effective methods to deal with the differences in ECG signals between the moving state and the resting state, which affects the practicality of the authentication system. Therefore, how to propose an ECG signal identity authentication method that can adapt to both the moving state and the resting state, and improve the accuracy and robustness of the authentication, is a problem that needs to be solved urgently. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and provide an identity authentication method based on mixed physiological state ECG signals, which can simultaneously adapt to and process ECG signals in motion and resting states, thereby improving the accuracy and robustness of identity authentication.

[0007] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] The present invention provides an identity authentication method based on mixed physiological state electrocardiogram signals, comprising:

[0009] Acquire mixed ECG signal data of the subject, wherein the mixed ECG signal data is obtained by mixing the ECG signals of the subject in a moving state and a resting state;

[0010] Preprocessing the mixed ECG signal data to obtain preprocessed mixed ECG signal data;

[0011] The preprocessed mixed ECG signal data is used as input, and the embedded feature vector is obtained based on the ECG signal extraction network model output;

[0012] The cosine similarity value is calculated based on the embedded feature vector, and the identity of the person to be tested is authenticated based on the cosine similarity value.

[0013] Further, the preprocessing of the motion-rest state mixed ECG signal data to obtain the preprocessed motion-rest state mixed ECG signal data includes:

[0014] A bandpass filter is used to filter the mixed ECG signal in the motion-rest state, wherein the bandpass filter is a fourth-order Butterworth bandpass filter with a frequency band range of 0.5-40 Hz;

[0015] A high-pass filter is used to eliminate the remaining baseline drift in the mixed ECG signal in the motion-rest state to obtain a filtered mixed ECG signal, wherein the high-pass filter is a first-order filter with a cutoff frequency set to 0.5 Hz;

[0016] The filtered mixed ECG signal is subjected to standardization processing to obtain a processed mixed ECG signal. The expression of the standardization processing is:

[0017]

[0018] in, represents the mixed ECG signal after filtering, represents the signal mean of the mixed ECG signal after filtering, represents the signal standard deviation of the filtered mixed ECG signal, represents the stability coefficient, represents the processed mixed ECG signal;

[0019] The processed mixed ECG signal is segmented and quality controlled to obtain preprocessed motion-rest state mixed ECG signal data.

[0020] Furthermore, the processed mixed ECG signal is segmented and quality controlled to obtain pre-processed motion-rest state mixed ECG signal data, including:

[0021] The Hamilton algorithm is used to detect the R wave position in the preprocessed mixed ECG signal and calculate the time interval between adjacent R waves;

[0022] Performing window segmentation at the R wave position to obtain a segmented signal, wherein the window division is centered on the R wave, 25% of the window length is intercepted forward, and the remaining part is intercepted backward, and the window does not exceed the effective range of the signal, wherein the window length is set according to the signal state;

[0023] Performing quality control detection on the segmented signals, screening out the segmented signals that meet the quality control conditions, i.e., serving as the pre-processed motion-rest state mixed ECG signal data;

[0024] The quality control conditions include the absence of non-digital values ​​and infinite values, the difference between the maximum signal value and the minimum signal value in the segmented signal is in the range of 1.0 to 15.0, and at least one valid R wave is included.

[0025] Furthermore, the ECG signal extraction network model includes a multi-scale convolution module, a feature enhancement module, a deep feature extraction module, a global context module and a self-attention module connected in sequence, and the output end of the self-attention module is connected in sequence to a global average pooling layer, a fully connected layer and a normalization layer;

[0026] The multi-scale convolution module is used to extract features from the pre-processed mixed ECG signal data to obtain a feature tensor, and the feature tensor is input into the feature enhancement module after feature fusion through channel dimension splicing;

[0027] The feature enhancement module is used to perform feature channel adaptive adjustment on the input feature tensor;

[0028] The deep feature extraction module is used to perform deep feature extraction on the input feature tensor;

[0029] The global context module is used to perform global information aggregation on the input features to obtain channel attention weights, and the channel attention weights are multiplied by the input feature tensor and then input into the self-attention module;

[0030] The self-attention module is used to establish long-range dependencies between input feature tensors.

[0031] Further, the feature enhancement module includes an adaptive average pooling layer, a first fully connected layer, a second fully connected layer and an output layer connected in sequence;

[0032] The adaptive average pooling layer is used to compress the input feature tensor into a one-dimensional feature vector;

[0033] The first fully connected layer is used to compress the channel dimension of the one-dimensional feature vector to 1 / 16 of the original dimension and process it through the ReLU activation function;

[0034] The second fully connected layer is used to restore the channel dimension of the one-dimensional feature vector to the original dimension, and obtain the attention weight coefficient of each channel through the Sigmoid activation function;

[0035] The output layer is used to multiply the weight coefficient of each channel by the input feature tensor channel by channel and then output it.

[0036] Furthermore, the multi-scale convolution module includes four parallel convolution branches, the convolution kernel sizes of the convolution branches are 3, 5, 7, and 11 respectively, and the number of input channels of each convolution branch is 1 and the number of output channels is 64;

[0037] The deep feature extraction module includes two sequentially connected convolution units, each of which is composed of a convolution layer, a batch normalization layer and a ReLU activation function, wherein the first convolution unit keeps the number of channels unchanged, and the second convolution unit increases the number of channels to 512;

[0038] The global context module includes an adaptive average pooling layer, two 1×1 convolutional layers and an activation function connected in sequence, wherein the first 1×1 convolutional layer reduces the number of channels to 256 and is processed by the ReLU activation function, and the second 1×1 convolutional layer restores the number of channels to 512 and is processed by the Sigmoid activation function to obtain the channel attention weight;

[0039] The self-attention module includes a Query branch, a Key branch and a Value branch. The Query branch and the Key branch reduce the number of channels to 1 / 8 of the input feature tensor through a 1×1 convolution layer, and the Value branch maintains the original number of channels through a 1×1 convolution layer. The product of the Query branch and the Key branch is processed by a Softmax activation function to obtain an attention score, which is then calculated with the Value branch to obtain an attention feature. The attention feature is enhanced by a learnable scaling parameter and a residual connection.

[0040] Furthermore, the output end of the normalization layer is also connected to a classifier, and the classifier is used to map features to a classification space for calculation of classification loss during the pre-training process of the ECG signal extraction network model.

[0041] Furthermore, it also includes pre-training the ECG signal extraction network model, and the pre-training method includes:

[0042] Obtain a mixed ECG signal dataset in motion and resting state;

[0043] The mixed ECG signal dataset in motion and resting state is divided into training set, validation set and test set;

[0044] The training set data is used as input to train the ECG signal extraction network model. During the training process, the Adam optimizer is used to optimize the parameters, adjust the learning rate, and calculate the loss function. Then the test set data is used as input to test the ECG signal extraction network model. The training is iterated continuously until the loss function during the test does not decrease for 10 consecutive rounds, and the training is stopped to obtain the trained ECG signal extraction network model.

[0045] A decision threshold is preset, and the validation set data is input into the trained ECG signal extraction network model to calculate the false acceptance rate and the false rejection rate. The false acceptance rate is the probability of misidentifying an unauthorized person as an authorized person, and the false rejection rate is the probability of misidentifying an authorized person as an unauthorized person. The decision threshold corresponding to the false acceptance rate and the false rejection rate being equal is the optimal decision threshold.

[0046] The decision threshold is set to the optimal decision threshold to obtain the pre-trained ECG signal extraction network model.

[0047] Furthermore, the loss function includes the sum of a focal loss function and a multiple similarity loss function, and the expression of the focal loss function is:

[0048]

[0049] in, is the category weight, is the probability of the predicted correct category, is the focus factor, is the cross entropy loss;

[0050] The multiple similarity loss function includes the similarity loss of positive sample pairs and the similarity loss of negative sample pairs, and its expression is:

[0051]

[0052]

[0053]

[0054] in, is the multiple similarity loss function, is the similarity loss of the positive sample pair, is the similarity loss of negative sample pairs, is the similarity of the positive sample pair, is the similarity of negative sample pairs, is the positive sample scaling factor, is the negative sample scaling factor, thresh is the similarity threshold, and clamp means limiting the input value to a preset range.

[0055] Furthermore, the calculating of the cosine similarity value between the data according to the embedded feature vector, and authenticating the identity of the subject to be tested according to the cosine similarity value between the data, includes:

[0056] The cosine similarity value between the data is calculated based on the embedded feature vector, and the expression of the cosine similarity value is:

[0057]

[0058] in, and All represent embedded feature vectors, n represents the dimension of embedded feature vectors, Represents the cosine similarity value;

[0059] The identity of the person being tested is authenticated based on the cosine similarity value between the data. If the cosine similarity value is greater than the optimal judgment threshold of the ECG signal extraction network model, the person being tested is judged as an authorized person. If the cosine similarity value is less than the optimal judgment threshold of the ECG signal extraction network model, the person being tested is judged as an unauthorized person.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The present invention realizes unified processing of ECG signals in different states by training with mixed motion state and rest state ECG data, and has good recognition performance for both motion state and rest state ECG signals, and has strong practical value; the ECG signal extraction network model captures ECG features of different scales through multi-scale feature extraction, adopts feature enhancement module to adaptively enhance important feature channels, and uses self-attention mechanism to establish long-range dependency between features, which significantly improves the effectiveness of feature extraction and improves the efficiency and accuracy of identity authentication;

[0062] The combined loss function training strategy proposed in the present invention solves the problem of sample category imbalance through focal loss and optimizes the feature space distribution using multi-similarity loss, which effectively improves the training effect and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A schematic diagram of a flow chart of an identity authentication method based on mixed physiological state electrocardiogram signals in an embodiment of the present invention;

[0064] Figure 2 A schematic diagram of the flow of pre-training and implementation of an identity authentication method based on mixed physiological state electrocardiogram signals in an embodiment of the present invention;

[0065] Figure 3 It is a structural schematic diagram of an ECG signal extraction network model of an identity authentication method based on mixed physiological state ECG signals in an embodiment of the present invention;

[0066] Figure 4 A schematic diagram of the structure of a feature enhancement module of an electrocardiogram signal extraction network model in an embodiment of the present invention;

[0067] Figure 5 It is a structural schematic diagram of a deep feature extraction module of an electrocardiogram signal extraction network model in an embodiment of the present invention;

[0068] Figure 6 A schematic diagram of the structure of a global context module of an electrocardiogram signal extraction network model in an embodiment of the present invention;

[0069] Figure 7 A schematic diagram of the structure of a self-attention module of an electrocardiogram signal extraction network model in one embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0071] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides an identity authentication method based on mixed physiological state ECG signals, which includes two stages, namely, an ECG registration stage and an ECG authentication stage, wherein the ECG registration stage is used to train the ECG signal extraction network model, and the ECG authentication stage is used to use the pre-trained ECG signal extraction network model to perform identity authentication on the subject.

[0072] First, the ECG registration phase specifically includes:

[0073] Construct an ECG signal extraction network model, such as Figure 3As shown, the ECG signal extraction network model includes a multi-scale convolution module, a feature enhancement module, a deep feature extraction module, a global context module and a self-attention module which are connected in sequence. The output end of the self-attention module is connected in sequence to a global average pooling layer, a fully connected layer and a normalization layer.

[0074] The multi-scale convolution module includes four parallel convolution branches, and the convolution kernel sizes of the convolution branches are 3, 5, 7, and 11 respectively. The number of input channels of each convolution branch is 1, and the number of output channels is 64. In the ECG signal, the QRS complex is a characteristic waveform that characterizes ventricular depolarization. It is composed of Q wave, R wave and S wave. It is the most important characteristic mark in ECG signal processing. The morphology, time interval and amplitude of the QRS complex can be used as an important basis for the identification of individual physiological characteristics. Smaller convolution kernels are used to capture local detail features such as R waves, and larger convolution kernels are used to capture morphological features such as QRS waveforms and relationship features between complexes.

[0075] The multi-scale convolution module is used to extract features from the preprocessed mixed ECG signal data through four convolution branches to obtain feature tensors, which are then concatenated and fused to obtain a 256-dimensional feature tensor that is input into the feature enhancement module.

[0076] Combination Figure 4 ,The feature enhancement module includes an adaptive average pooling layer, a first fully connected layer, a second fully connected layer and an output layer which are connected in sequence.

[0077] The processing process of the feature enhancement module is divided into two processing stages: feature compression and adaptive recalibration. In the feature compression stage, the input feature tensor is compressed into a one-dimensional feature vector through the adaptive average pooling layer. In the adaptive recalibration stage, the first fully connected layer and the second fully connected layer in series are used to learn the dependency between channels. Among them, the first fully connected layer compresses the channel dimension to 1 / 16 of the original dimension and is processed by the ReLU activation function. The second fully connected layer restores the channel dimension to the original dimension and is processed by the Sigmoid activation function to obtain the weight coefficient of each channel. Finally, the weight coefficient of each channel is multiplied by the input feature tensor channel by channel through the output layer and output to the deep feature extraction module.

[0078] Combination Figure 5 The deep feature extraction module is used to further extract deep features. It includes two convolutional units connected in sequence. Each convolutional unit consists of a convolutional layer, a batch normalization layer and a ReLU activation function. The first convolutional unit keeps the number of channels unchanged, and the second convolutional unit increases the number of channels to 512 dimensions.

[0079] Combination Figure 6The global context module includes an adaptive average pooling layer, two 1×1 convolutional layers and activation functions connected in sequence. The first 1×1 convolutional layer reduces the channel to 256 and processes it through the ReLU activation function. The second 1×1 convolutional layer restores the channel to 512 and processes it through the Sigmoid activation function to obtain the channel attention weight. The channel attention weight is multiplied by the input feature tensor and output to the self-attention module.

[0080] Combination Figure 7 The self-attention module establishes long-range dependencies between features through the Query-Key-Value mechanism. The Query and Key branches reduce the number of channels to 1 / 8 of the original through 1×1 convolution, and the Value branch maintains the original number of channels. The product of Query and Key is activated by Softmax to obtain the attention score, and the attention feature is calculated with the Value branch. Finally, the enhanced feature representation is obtained through the learnable scaling parameter and residual connection, and the attention strength is adjusted by introducing the learnable scaling parameter γ.

[0081] The enhanced feature representation is subjected to global average pooling through a global average pooling layer to obtain a 512-dimensional feature tensor. The feature dimension is reduced to 64 dimensions through a fully connected layer, and L2 normalization is performed to obtain an embedded feature vector.

[0082] In this embodiment, in order to pre-train the ECG signal extraction network model, the output end of the normalization layer is also connected to a classifier, which is used to map features to a classification space for calculating classification losses during the pre-training process of the ECG signal extraction network model.

[0083] The constructed ECG signal extraction network model is pre-trained, and the pre-training method specifically includes:

[0084] Obtain a mixed ECG signal dataset in motion and resting states.

[0085] The mixed ECG signal dataset in motion and resting state is divided into training set, validation set and test set.

[0086] The training set data is used as input to train the ECG signal extraction network model. During the training process, the Adam optimizer is used to optimize the parameters, adjust the learning rate, and calculate the loss function. Then the test set data is used as input to test the ECG signal extraction network model. The training is iterated continuously until the loss function during the test does not decrease for 10 consecutive rounds, and the training is stopped to obtain the trained ECG signal extraction network model.

[0087] The loss function is the sum of the focal loss function and the multiple similarity loss function. The expression of the focal loss function is:

[0088]

[0089] in, is the category weight, is the probability of the predicted correct category, is the focus factor, which controls the attention paid to different samples. is the cross entropy loss.

[0090] The multiple similarity loss function includes the similarity loss of positive sample pairs and the similarity loss of negative sample pairs, and the expression is:

[0091]

[0092]

[0093]

[0094] in, is the similarity of positive sample pairs, is the similarity of negative sample pairs, is the positive sample scaling factor, is the negative sample scaling factor, thresh is the similarity threshold, and clamp means limiting the input value to a certain range to ensure numerical stability.

[0095] A decision threshold is preset, and the validation set data is input into the trained ECG signal extraction network model to calculate the false acceptance rate and the false rejection rate. The false acceptance rate is the probability of misidentifying an unauthorized person as an authorized person, and the false rejection rate is the probability of misidentifying an authorized person as an unauthorized person. The decision threshold corresponding to the false acceptance rate and the false rejection rate being equal is the optimal decision threshold.

[0096] The decision threshold is set to the optimal decision threshold to obtain the pre-trained ECG signal extraction network model.

[0097] Next, we will enter the ECG certification stage, which includes:

[0098] The mixed ECG signal data of the subject is obtained by mixing the ECG signals of the subject in a moving state and a resting state.

[0099] Preprocess the mixed ECG signal data, specifically:

[0100] The signal is filtered using a fourth-order Butterworth bandpass filter. The frequency band of the filter is set to 0.5Hz-40Hz. The reason for selecting this frequency band is that the main effective components of the ECG signal are concentrated in this range, which can effectively filter out power frequency interference (50Hz / 60Hz) and baseline drift (<0.5Hz).

[0101] A high-pass filter with a cutoff frequency of 0.5 Hz is used to further eliminate the remaining baseline drift in the signal. The high-pass filter uses a first-order filter to remove the baseline drift while maintaining the original shape of the signal as much as possible.

[0102] The filtered mixed ECG signal is standardized to obtain the processed mixed ECG signal. The expression of the standardized processing is:

[0103]

[0104] in, represents the mixed ECG signal after filtering, represents the signal mean of the mixed ECG signal after filtering, represents the signal standard deviation of the filtered mixed ECG signal, represents the stability coefficient, Represents the processed mixed ECG signal.

[0105] Segment the processed mixed ECG signal: Use Hamilton algorithm to detect the R wave position in the preprocessed mixed ECG signal, and calculate the time interval between adjacent R waves.

[0106] Window segmentation is performed at the R wave position to obtain a segmented signal, wherein the window division is centered on the R wave, 25% of the window length is intercepted forward, and the remaining part is intercepted backward, and the window does not exceed the effective range of the signal, wherein the window length is set according to the signal state.

[0107] The segmented signals were subjected to quality control testing to screen out those that met the quality control conditions. The quality control conditions included the absence of non-numeric values ​​and infinite values, the difference between the maximum and minimum values ​​of the segmented signals being in the range of 1.0 to 15.0, and containing at least one valid R wave. The obtained segmented signals were used as the preprocessed motion-rest state mixed ECG signal data.

[0108] The preprocessed motion-rest state mixed ECG signal data is used as input, and a pre-trained ECG signal extraction network model is used to obtain multiple embedded feature vectors, each of which corresponds to a segmented signal data.

[0109] Calculate the cosine similarity value of the pairwise embedded feature vectors. The calculation expression is:

[0110]

[0111] in, and Both represent embedded feature vectors. The cosine similarity value ranges from [-1,1]. The closer the value is to 1, the more similar the two feature vectors are, and the closer it is to -1, the less similar they are.

[0112] The identity of the person to be tested is authenticated according to the cosine similarity value. If the cosine similarity value is greater than the optimal judgment threshold of the ECG signal extraction network model, the person to be tested is determined to be an authorized person. If the cosine similarity value is less than the optimal judgment threshold of the ECG signal extraction network model, the person to be tested is determined to be an unauthorized person.

[0113] The present invention has been fully experimentally verified on a mixed-state ECG data set. The optimal threshold determined on the validation set is 0.6600. This threshold is used to perform authentication evaluation on the test set, and excellent performance is achieved, with an authentication accuracy of 93.57%, a false acceptance rate of 1.97%, and a false rejection rate of 10.9%, indicating that the method of the present invention has strong robustness and generalization ability. These experimental results also fully demonstrate the effectiveness and practical value of the present invention in processing mixed motion-resting state ECG data.

[0114] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An identity authentication method based on mixed physiological state electrocardiogram signals, characterized in that: include: Acquire mixed ECG signal data of the subject, wherein the mixed ECG signal data is obtained by mixing the ECG signals of the subject in a moving state and a resting state; Preprocessing the mixed ECG signal data to obtain preprocessed mixed ECG signal data; The preprocessed mixed ECG signal data is used as input, and the embedded feature vector is obtained based on the ECG signal extraction network model output; The cosine similarity value is calculated based on the embedded feature vector, and the identity of the person to be tested is authenticated based on the cosine similarity value.

2. The identity authentication method based on mixed physiological state electrocardiogram signals according to claim 1, characterized in that: The preprocessing of the motion-rest state mixed ECG signal data to obtain the preprocessed motion-rest state mixed ECG signal data includes: A bandpass filter is used to filter the mixed ECG signal in the motion-rest state, wherein the bandpass filter is a fourth-order Butterworth bandpass filter with a frequency band range of 0.5-40 Hz; A high-pass filter is used to eliminate the remaining baseline drift in the mixed ECG signal in the motion-rest state to obtain a filtered mixed ECG signal, wherein the high-pass filter is a first-order filter with a cutoff frequency set to 0.5 Hz; The filtered mixed ECG signal is subjected to standardization processing to obtain a processed mixed ECG signal. The expression of the standardization processing is: ； in, represents the mixed ECG signal after filtering, represents the signal mean of the mixed ECG signal after filtering, represents the signal standard deviation of the filtered mixed ECG signal, represents the stability coefficient, represents the processed mixed ECG signal; The processed mixed ECG signal is segmented and quality controlled to obtain preprocessed motion-rest state mixed ECG signal data.

3. The identity authentication method based on mixed physiological state electrocardiogram signal according to claim 2, characterized in that: The processed mixed ECG signal is segmented and quality controlled to obtain pre-processed motion-rest state mixed ECG signal data, including: The Hamilton algorithm is used to detect the R wave position in the preprocessed mixed ECG signal and calculate the time interval between adjacent R waves; Performing window segmentation at the R wave position to obtain a segmented signal, wherein the window division is centered on the R wave, 25% of the window length is intercepted forward, and the remaining part is intercepted backward, and the window does not exceed the effective range of the signal, wherein the window length is set according to the signal state; Performing quality control detection on the segmented signals, screening out the segmented signals that meet the quality control conditions, i.e., serving as the pre-processed motion-rest state mixed ECG signal data; The quality control conditions include the absence of non-digital values ​​and infinite values, the difference between the maximum signal value and the minimum signal value in the segmented signal is in the range of 1.0 to 15.0, and at least one valid R wave is included.

4. The identity authentication method based on mixed physiological state electrocardiogram signal according to claim 1, characterized in that: The ECG signal extraction network model includes a multi-scale convolution module, a feature enhancement module, a deep feature extraction module, a global context module and a self-attention module which are connected in sequence, and the output end of the self-attention module is connected in sequence to a global average pooling layer, a fully connected layer and a normalization layer; The multi-scale convolution module is used to extract features from the pre-processed mixed ECG signal data to obtain a feature tensor, and the feature tensor is input into the feature enhancement module after feature fusion through channel dimension splicing; The feature enhancement module is used to perform feature channel adaptive adjustment on the input feature tensor; The deep feature extraction module is used to perform deep feature extraction on the input feature tensor; The global context module is used to perform global information aggregation on the input features to obtain channel attention weights, and the channel attention weights are multiplied by the input feature tensor and then input into the self-attention module; The self-attention module is used to establish long-range dependencies between input feature tensors.

5. The identity authentication method based on mixed physiological state electrocardiogram signal according to claim 4, characterized in that: The feature enhancement module includes an adaptive average pooling layer, a first fully connected layer, a second fully connected layer and an output layer connected in sequence; The adaptive average pooling layer is used to compress the input feature tensor into a one-dimensional feature vector; The first fully connected layer is used to compress the channel dimension of the one-dimensional feature vector to 1 / 16 of the original dimension and process it through the ReLU activation function; The second fully connected layer is used to restore the channel dimension of the one-dimensional feature vector to the original dimension, and obtain the attention weight coefficient of each channel through the Sigmoid activation function; The output layer is used to multiply the weight coefficient of each channel by the input feature tensor channel by channel and then output it.

6. The identity authentication method based on mixed physiological state electrocardiogram signal according to claim 4, characterized in that: The multi-scale convolution module includes four parallel convolution branches, the convolution kernel sizes of the convolution branches are 3, 5, 7, and 11 respectively, and the number of input channels of each convolution branch is 1 and the number of output channels is 64; The deep feature extraction module includes two sequentially connected convolution units, each of which is composed of a convolution layer, a batch normalization layer and a ReLU activation function, wherein the first convolution unit keeps the number of channels unchanged, and the second convolution unit increases the number of channels to 512; The global context module includes an adaptive average pooling layer, two 1×1 convolutional layers and an activation function connected in sequence, wherein the first 1×1 convolutional layer reduces the number of channels to 256 and is processed by the ReLU activation function, and the second 1×1 convolutional layer restores the number of channels to 512 and is processed by the Sigmoid activation function to obtain the channel attention weight; The self-attention module includes a Query branch, a Key branch and a Value branch. The Query branch and the Key branch reduce the number of channels to 1 / 8 of the input feature tensor through a 1×1 convolution layer, and the Value branch maintains the original number of channels through a 1×1 convolution layer. The product of the Query branch and the Key branch is processed by a Softmax activation function to obtain an attention score, which is then calculated with the Value branch to obtain an attention feature. The attention feature is enhanced by a learnable scaling parameter and a residual connection.

7. The identity authentication method based on mixed physiological state electrocardiogram signals according to claim 4, characterized in that: The output end of the normalization layer is also connected to a classifier, which is used to map features to a classification space for calculation of classification loss during the pre-training process of the electrocardiogram signal extraction network model.

8. The identity authentication method based on mixed physiological state electrocardiogram signal according to claim 1, characterized in that: The method also includes pre-training the ECG signal extraction network model, wherein the pre-training method includes: Obtain a mixed ECG signal dataset in motion and resting state; The mixed ECG signal dataset in motion and resting state is divided into training set, validation set and test set; The training set data is used as input to train the ECG signal extraction network model. During the training process, the Adam optimizer is used to optimize the parameters, adjust the learning rate, and calculate the loss function. Then the test set data is used as input to test the ECG signal extraction network model. The training is iterated continuously until the loss function during the test does not decrease for 10 consecutive rounds, and the training is stopped to obtain the trained ECG signal extraction network model. A decision threshold is preset, and the validation set data is input into the trained ECG signal extraction network model to calculate the false acceptance rate and the false rejection rate. The false acceptance rate is the probability of misidentifying an unauthorized person as an authorized person, and the false rejection rate is the probability of misidentifying an authorized person as an unauthorized person. The decision threshold corresponding to the false acceptance rate and the false rejection rate being equal is the optimal decision threshold. The decision threshold is set to the optimal decision threshold to obtain the pre-trained ECG signal extraction network model.

9. The identity authentication method based on mixed physiological state electrocardiogram signals according to claim 7, characterized in that: The loss function includes the sum of the focus loss function and the multiple similarity loss function. The expression of the focus loss function is: ； in, is the category weight, is the probability of the predicted correct category, is the focus factor, is the cross entropy loss; The multiple similarity loss function includes the similarity loss of positive sample pairs and the similarity loss of negative sample pairs, and its expression is: ； ； ； in, is the multiple similarity loss function, is the similarity loss of the positive sample pair, is the similarity loss of negative sample pairs, is the similarity of the positive sample pair, is the similarity of negative sample pairs, is the positive sample scaling factor, is the negative sample scaling factor, thresh is the similarity threshold, and clamp means limiting the input value to a preset range.

10. The identity authentication method based on mixed physiological state electrocardiogram signal according to claim 1, characterized in that: The step of calculating the cosine similarity value between the data according to the embedded feature vector and authenticating the identity of the person to be tested according to the cosine similarity value between the data includes: The cosine similarity value between the data is calculated based on the embedded feature vector, and the expression of the cosine similarity value is: ； in, and All represent embedded feature vectors, n represents the dimension of embedded feature vectors, Represents the cosine similarity value; The identity of the person being tested is authenticated based on the cosine similarity value between the data. If the cosine similarity value is greater than the optimal judgment threshold of the ECG signal extraction network model, the person being tested is judged as an authorized person. If the cosine similarity value is less than the optimal judgment threshold of the ECG signal extraction network model, the person being tested is judged as an unauthorized person.

Images