Identity recognition method, device and equipment based on photoelectric volume pulse wave and medium

Abstract

The application relates to an identity recognition method based on a photoelectric volume pulse wave, which comprises the following steps: obtaining target PPG signal data; obtaining a target classification recognition result corresponding to the target PPG signal data according to the target PPG signal data and a target identity recognition model; the target identity recognition model is obtained by training a historical identity recognition model based on a current PPG signal database; and the target identity recognition model comprises an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted in the process of training the target identity recognition model based on the historical identity recognition model. The method can maintain high classification recognition accuracy when the input PPG signal database changes.

Description

Technical Field

[0001] This application relates to the field of biosignal recognition technology, and in particular to an identity recognition method, device, equipment and medium based on photoplethysmography pulse waves. Background Technology

[0002] PPG (Photo Plethysmo Graphy) signal is a medical physiological signal with individual specificity, and identification methods based on PPG signals are currently widely used identification methods.

[0003] In related technologies, identity recognition methods based on PPG signals generally include: first, training a convolutional neural network for identity recognition based on the current PPG signal database; then, using the convolutional neural network to extract features and classify the acquired PPG signals to obtain the user identity corresponding to the PPG signal.

[0004] However, the aforementioned PPG signal-based identification method has low accuracy when applied to identification systems with high personnel mobility. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, device, equipment, and medium for identity recognition based on photoplethysmography (PPG) waves that can achieve high accuracy in identity recognition systems with high personnel mobility, addressing the aforementioned technical problems.

[0006] In a first aspect, this application provides an identity recognition method based on photoplethysmography (PPG), the method comprising:

[0007] Acquire target PPG signal data;

[0008] Based on the target PPG signal data and the target identity recognition model, the target classification and recognition results corresponding to the target PPG signal data are obtained; the target identity recognition model is obtained by training the historical identity recognition model based on the current PPG signal database; the target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted in the process of training the target identity recognition model based on the historical identity recognition model.

[0009] In one embodiment, the method further includes:

[0010] The initial identity recognition model is iteratively trained based on the historical PPG signal database to obtain the historical identity recognition model and its recognition accuracy.

[0011] The weight matrix is ​​adjusted based on the recognition accuracy of the historical identity recognition model.

[0012] In one embodiment, the initial identity recognition model is iteratively trained based on a historical PPG signal database to obtain a historical identity recognition model and the recognition accuracy of the historical identity recognition model, including:

[0013] Obtain the historical PPG signal database, which includes multiple historical PPG signal data and the historical user identity corresponding to each historical PPG signal data;

[0014] For each historical PPG signal data, the historical PPG signal data is input into the initial identity recognition model to obtain the initial recognition result output by the initial identity recognition model;

[0015] The initial identity recognition model is iteratively trained according to the preset iteration conditions to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model.

[0016] In one embodiment, the weight matrix is ​​adjusted based on the recognition accuracy of the historical identity recognition model, including:

[0017] Based on the recognition accuracy, the gradient of each model parameter in the historical identity model is calculated to obtain the gradient matrix of the first parameter.

[0018] Transpose the gradient matrix of the first parameter to obtain the gradient matrix of the second parameter;

[0019] The inner product of the gradients of the first and second parameters is calculated to obtain the adjusted weight matrix.

[0020] In one embodiment, the method further includes:

[0021] Obtain the current PPG signal database and historical identity recognition model; the current PPG signal database includes multiple current PPG signal data and the current user identity corresponding to each current PPG signal data.

[0022] For each current PPG signal data, the historical identity recognition model is used to classify and identify the current PPG signal data to obtain intermediate recognition results;

[0023] Based on the intermediate identification results, the current user identity corresponding to the current PPG signal data, and the loss function, the parameters of each model in the historical identity recognition model are adjusted to obtain the intermediate identity recognition model; the regularization term of the loss function is determined based on the adjusted weight matrix and the model parameters of the historical identity recognition model.

[0024] The intermediate identity recognition model is trained iteratively until a preset iteration condition is met. Then, the intermediate identity recognition model corresponding to the preset iteration condition is output to obtain the target identity recognition model.

[0025] In one embodiment, based on the intermediate identification results, the current user identity corresponding to the current PPG signal data, and the loss function, the parameters of each model in the historical identity recognition model are adjusted to obtain an intermediate identity recognition model, including:

[0026] For each weight data in the weight adjustment matrix, if the weight data is less than a preset threshold, the model parameters corresponding to the weight data in the historical identity recognition model are adjusted according to the loss function to obtain an intermediate identity recognition model.

[0027] If the weight data is greater than or equal to the preset threshold, the model parameters corresponding to the weight data in the historical identity recognition model remain unchanged, and an intermediate identity recognition model is obtained.

[0028] The weight data corresponds to the model parameters of the historical identity recognition model.

[0029] Secondly, this application also provides an identity recognition device based on photoplethysmography (PPG), the device comprising:

[0030] The data acquisition module is used to acquire target PPG signal data;

[0031] The classification and recognition module is used to obtain the target classification and recognition result corresponding to the target PPG signal data based on the target PPG signal data and the target identity recognition model. The target identity recognition model is obtained by training the historical identity recognition model based on the current PPG signal database. The target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted in the process of training the target identity recognition model based on the historical identity recognition model.

[0032] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in the first aspect.

[0033] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0034] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect.

[0035] The aforementioned photoplethysmography (PPG)-based identity recognition method, apparatus, device, and medium acquire target PPG signal data; based on the target PPG signal data and the target identity recognition model, obtain the target classification and recognition result corresponding to the target PPG signal data; the target identity recognition model is obtained by training historical identity recognition models based on the current PPG signal database; the target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted during the training of the target identity recognition model based on the historical identity recognition model; thus, the parameters of the target identity recognition model are adjusted according to different adjustment weights based on the model parameters of the historical identity recognition model and the adjustment weight matrix, avoiding the interference of new information on the knowledge learned by the original identity recognition model when the input database changes in traditional transfer learning-based identity recognition models. Furthermore, since the model parameters of the original identity recognition model share weights, the interference information can cause catastrophic forgetting problems in the identity recognition model. In this application, different adjustment weights are set for different model parameters, so that during the training of the target identity recognition model, the model parameters of the historical identity recognition model are adjusted within different ranges, and the resulting target identity recognition model can maintain high classification accuracy when the input PPG signal database changes. Attached Figure Description

[0036] Figure 1 This is an application environment diagram of an identity recognition method based on photoplethysmography (PPG) in one embodiment.

[0037] Figure 2 This is a flowchart illustrating an identity recognition method based on photoplethysmography (PPG) in one embodiment.

[0038] Figure 3 This is a flowchart illustrating the steps for obtaining the adjusted weight matrix in one embodiment;

[0039] Figure 4 This is a flowchart illustrating the steps for obtaining the historical identity recognition model and recognition accuracy in one embodiment;

[0040] Figure 5 This is a flowchart illustrating the steps for obtaining the adjusted weight matrix in one embodiment;

[0041] Figure 6 This is a flowchart illustrating the steps for obtaining a target identity recognition model in one embodiment;

[0042] Figure 7 This is a flowchart illustrating the steps for obtaining an intermediate identity recognition model in one embodiment;

[0043] Figure 8This is a flowchart illustrating an identity recognition method based on photoplethysmography (PPG) waves in another embodiment.

[0044] Figure 9 This is a structural block diagram of an identity recognition device based on photoplethysmography (PPG) in one embodiment.

[0045] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] The identity recognition method based on photoplethysmography provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0048] Terminal 102 acquires target PPG signal data; based on the target PPG signal data and the target identity recognition model, it obtains the target classification and recognition result corresponding to the target PPG signal data; the target identity recognition model is obtained by training the historical identity recognition model based on the current PPG signal database; the target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted in the process of training the target identity recognition model based on the historical identity recognition model.

[0049] In one embodiment, such as Figure 2 As shown, an identity recognition method based on photoplethysmography (PPG) is provided, which is then applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps:

[0050] Step 202: Acquire target PPG signal data.

[0051] The target PPG signal data is obtained by preprocessing the directly acquired initial PPG signal in a preset manner.

[0052] Step 204: Based on the target PPG signal data and the target identity recognition model, obtain the target classification and recognition result corresponding to the target PPG signal data.

[0053] The target identity recognition model is obtained by training historical identity recognition models based on the current PPG signal database; the target identity recognition model includes an adjusted weight matrix.

[0054] Among them, the historical identity recognition model is trained based on historical PPG signals and the historical user identities corresponding to historical PPG signals in the historical PPG signal database.

[0055] In the process of training the target identity recognition model, the current PPG signal and the current user identity corresponding to the current PPG signal in the current PPG signal database are input into the historical identity recognition model for training. The model obtained after meeting the preset conditions is the target identity recognition model.

[0056] For example, the historical identity recognition model can be a deep learning model such as a convolutional neural network, a recurrent neural network, or a deep belief network. The historical identity recognition model includes a feature extraction layer and a decision layer. The current PPG signal database is input into the historical identity recognition model for training. The model obtained after meeting preset conditions is the target identity recognition model. When the sizes of the historical PPG signal database and the current PPG signal database are different, the decision layer parameters of the historical identity recognition model need to be adaptively adjusted first, and then the historical identity recognition model is trained based on the current PPG signal database.

[0057] The weight adjustment matrix is ​​used to determine the model parameters that need to be adjusted during the process of training the target identity recognition model based on the historical identity recognition model.

[0058] The weighting matrix was adjusted to evaluate the importance of each model parameter in the historical identity recognition model. Each weight in the weighting matrix represents the importance of its corresponding model parameter. During the training of the target identity recognition model based on the historical model, the weighting matrix constrains the adjustment of each model parameter within different ranges according to their varying importance. This ensures that the trained target identity recognition model retains the knowledge learned from the historical PPG signal database.

[0059] For example, the weight adjustment matrix can be the second-order partial derivative matrix of each model parameter of the historical identity recognition model with respect to the loss function, or it can be defined using a first-order method.

[0060] In the aforementioned photoplethysmography (PPG)-based identity recognition method, target PPG signal data is acquired; based on the target PPG signal data and the target identity recognition model, the target classification and recognition result corresponding to the target PPG signal data is obtained; the target identity recognition model is trained on historical identity recognition models using the current PPG signal database; the target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted during the training process of the target identity recognition model based on the historical identity recognition model; thus, the parameters of the target identity recognition model are adjusted according to different adjustment weights based on the model parameters of the historical identity recognition model and the adjustment weight matrix, avoiding the interference of new information on the knowledge learned by the original identity recognition model when the input database changes, which is common in traditional transfer learning-based identity recognition models. Furthermore, because the model parameters of the original identity recognition model share weights, interference information can cause catastrophic forgetting problems in the identity recognition model. In this application, different adjustment weights are set for different model parameters, so that during the training process of the target identity recognition model, the model parameters of the historical identity recognition model are adjusted within different ranges, and the resulting target identity recognition model can maintain high classification and recognition accuracy when the input PPG signal database changes.

[0061] In one embodiment, based on Figure 2 The illustrated embodiments, such as Figure 3 As shown, the method also includes:

[0062] Step 302: Iteratively train the initial identity recognition model based on the historical PPG signal database to obtain the historical identity recognition model and its recognition accuracy.

[0063] For example, the initial identity recognition model can be a deep learning model such as a convolutional neural network, a recurrent neural network, or a deep belief network.

[0064] The initial identity recognition model is iteratively trained, and the model that reaches the preset iteration conditions is used as the historical identity recognition model. The accuracy of the historical identity recognition model based on the recognition results of the historical PPG signal database is used as the recognition precision of the historical identity recognition model.

[0065] Step 304: Obtain the adjusted weight matrix based on the recognition accuracy of the historical identity recognition model.

[0066] In this embodiment, the weight matrix is ​​adjusted by utilizing the recognition accuracy of the historical identity recognition model. The recognition results based on the historical PPG signal database are used as constraints in the training process of the target identity recognition model. This enables the target identity recognition model to maintain its memory of the historical PPG signal database when learning the new task of the current PPG signal database, thereby improving the stability of the target identity recognition model in classification and recognition when the input database changes.

[0067] In one embodiment, based on Figure 3 The illustrated embodiments, such as Figure 4 As shown in this embodiment, the process of iteratively training an initial identity recognition model based on a historical PPG signal database to obtain a historical identity recognition model and the recognition accuracy of the historical identity recognition model includes:

[0068] Step 402: Obtain the historical PPG signal database.

[0069] The historical PPG signal database includes multiple historical PPG signal datasets and the corresponding historical user identities for each historical PPG signal dataset. The historical PPG signal data is obtained by preprocessing raw PPG signals directly collected from various historical users in a uniform manner, resulting in PPG signal data with similar edge and geometric features.

[0070] Step 404: For each historical PPG signal data, input the historical PPG signal data into the initial identity recognition model to obtain the initial recognition result output by the initial identity recognition model.

[0071] The initial identification result can be either successful, meaning that the historical PPG signal data corresponds to a historical user identity in the historical PPG signal database, or unsuccessful, meaning that the historical PPG signal data does not correspond to any historical user identity in the historical PPG signal database.

[0072] Step 406: Iteratively train the initial identity recognition model according to the preset iteration conditions to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model.

[0073] After each training iteration, the model parameters of the initial identity recognition model are adjusted based on the recognition results until the preset iteration conditions are met. The historical identity recognition model and the recognition accuracy of the historical identity recognition model when the preset iteration conditions are met are then output.

[0074] In this embodiment, the initial identity recognition model is iteratively trained based on the historical PPG signal database to obtain the historical identity recognition model and its recognition accuracy. The historical identity recognition model is used to train the target identity recognition model, and the recognition accuracy of the historical identity recognition model is used as the basis for calculating the adjustment weight matrix. This ensures that the constraints on the model parameters of the target identity recognition model and the historical identity recognition model can maintain the stability of the target identity recognition model in classification and recognition based on the historical PPG signal database.

[0075] In one embodiment, based on Figure 3 The illustrated embodiments, such as Figure 5 As shown in this embodiment, the process of obtaining the adjusted weight matrix based on the historical recognition accuracy of the historical identity recognition model includes:

[0076] Step 502: Calculate the gradient of each model parameter in the historical identity model based on the historical recognition accuracy to obtain the gradient of the first parameter.

[0077] Here, the first parameter gradient is the matrix of prior probability gradients obtained by calculating the gradients of the historical recognition accuracy with respect to the parameters of each model in the historical identity model.

[0078] Step 504: Transpose the first parameter gradient matrix to obtain the second parameter gradient matrix.

[0079] Step 506: Calculate the inner product of the gradients of the first and second parameters to obtain the adjusted weight matrix.

[0080] For example, adjusting the weight matrix F can be represented as the gradient of the first parameter. Second parameter gradient The inner product of can be expressed as:

[0081]

[0082] Where, x n This represents historical PPG signal data in the historical PPG signal database, y n This represents the historical PPG signal data that was correctly classified in the historical identification results. These are the model parameters in the historical identity model.

[0083] In this embodiment, the weight matrix is ​​adjusted based on the historical recognition accuracy of the historical identity recognition model. The importance of each parameter in the historical identity recognition model is determined by the historical recognition accuracy. Based on the importance of the parameter, it is determined whether the parameter needs to be adjusted during the training of the target identity recognition model, thereby improving the stability of the target identity recognition model in classification and recognition when the input database changes.

[0084] In one embodiment, based on Figure 3 The illustrated embodiments, such as Figure 6 As shown, the method provided in this embodiment further includes:

[0085] Step 602: Obtain the current PPG signal database and historical identity recognition model.

[0086] The current PPG signal database includes multiple current PPG signal data sets and the target user identities corresponding to each target PPG signal data set. The current PPG signal data sets are obtained by preprocessing raw PPG signals directly collected from each current user in a uniform manner, resulting in PPG signal data sets with similar edge and geometric features.

[0087] Step 604: For each target PPG signal data, the historical identity recognition model is used to classify and identify the target PPG signal data to obtain intermediate recognition results.

[0088] The intermediate identification result can be either successful, meaning that the target PPG signal data corresponds to a target user identity in the current PPG signal database, or unsuccessful, meaning that the target PPG signal data does not correspond to any target user identity in the current PPG signal database.

[0089] For example, the historical identity recognition model can be a convolutional neural network, including a feature extraction layer and a decision layer. The target PPG signal data is input into the historical identity recognition model, and features are extracted through the feature extraction layer. Then, the extracted features are input into the decision layer for classification and recognition, and an intermediate recognition result of recognition success or failure is obtained.

[0090] Step 606: Based on the intermediate identification results, the target user identity corresponding to the target PPG signal data, and the loss function, adjust the parameters of each model in the historical identity recognition model to obtain the intermediate identity recognition model.

[0091] The loss function includes a basic loss term and a regularization term. The basic loss term can be obtained based on the user identity corresponding to the intermediate recognition results and PPG signal data. The regularization term can be determined based on the adjusted weight matrix and the model parameters of the historical identity recognition model.

[0092] For example, the loss function L B It can be represented as:

[0093]

[0094] Where, L(θ) iL(θ) represents the basic loss term of the loss function, which can be a function of cross-entropy loss, perceptual loss, or mean squared error loss, etc. λ is a weight adjustment factor. L(θ) and λ can be selected according to the application scenario, and this application does not impose any restrictions on them; i This represents the i-th model parameter of the identity recognition model that is performing the current classification and recognition process.

[0095] The parameters of each model in the historical identity recognition model are adjusted based on the value of the loss function to obtain the intermediate identity recognition model.

[0096] Step 608: Iteratively train the intermediate identity recognition model until the preset iteration conditions are met, and output the intermediate identity recognition model corresponding to the preset iteration conditions to obtain the target identity recognition model.

[0097] During each iteration of training, the regularization term of the loss function is determined based on the model parameters and adjusted weight matrix of the intermediate identity recognition model in the current iteration process. The basic loss term of the loss function is obtained based on the user identity corresponding to the intermediate recognition results and PPG signal data, thereby obtaining the value of the loss function for the current iteration process. The model parameters of the intermediate identity recognition model in the current iteration process are adjusted based on the value of the loss function to obtain the intermediate identity recognition model for the next iteration process, until the preset iteration conditions are met.

[0098] In this embodiment, based on the intermediate identification results, the target user identity corresponding to the target PPG signal data, and the loss function, the model parameters of the historical identity recognition model and the intermediate identity recognition model are adjusted during iterative training to obtain a target identity recognition model that meets the preset iteration conditions. Thus, the weight adjustment matrix, as part of the regularization term of the loss function, influences the adjustment of the model parameters of the intermediate identity recognition model during each iterative training process. The weight adjustment matrix allows for controllable adjustment of the target identity recognition model's parameters when the PPG signal database input to the target identity recognition model changes, thereby maintaining stable classification accuracy. This embodiment trains based on the historical identity recognition model and the current PPG signal database. When there is overlap between the current and historical PPG signal databases, the training time for the overlapping portion can be saved, thereby shortening the training time of the target identity recognition model and improving its training efficiency.

[0099] In one embodiment, based on Figure 6 The illustrated embodiments, such as Figure 7 As shown in this embodiment, the process of adjusting the parameters of the historical identity recognition model based on the intermediate recognition results, the target user identity corresponding to the target PPG signal data, and the loss function to obtain the intermediate identity recognition model includes:

[0100] Step 702: For each weight data in the adjusted weight matrix, determine whether the weight data is less than a preset threshold.

[0101] The weight data corresponds to the model parameters of the historical identity recognition model.

[0102] Step 704: If the weight data is less than the preset threshold, the model parameters of the historical identity recognition model are adjusted according to the loss function to obtain the intermediate identity recognition model.

[0103] When the weight data is less than the preset threshold, it means that the model parameters of the historical identity recognition model corresponding to the weight data are less important. After adjusting the parameters, the intermediate identity recognition model obtained by recognizing historical PPG signals in the historical PPG signal database has a smaller difference from the historical recognition results.

[0104] Step 706: If the weight data is greater than or equal to the preset threshold, then keep the model parameters of the initial identity recognition model unchanged to obtain the intermediate identity recognition model.

[0105] When the weight data is greater than or equal to the preset threshold, it indicates that the model parameters of the historical identity recognition model corresponding to the weight data are of high importance. If the parameter is adjusted, the intermediate recognition result of the intermediate identity recognition model will be significantly different from the historical recognition result, thereby affecting the target recognition accuracy of the target identity recognition model.

[0106] In this embodiment, the model parameters that need to be adjusted in the historical identity recognition model are determined based on the weight data. In this way, during the process of iteratively training the target identity recognition model based on the historical identity model, the parameters that have a smaller impact on the historical PPG signal database are selectively updated, thereby improving the recognition accuracy of the target identity recognition model in the classification and recognition task based on the highly mobile PPG signal database.

[0107] In one embodiment, such as Figure 8 As shown, an identity recognition method based on photoplethysmography (PPG) is provided, the method comprising:

[0108] Step 802: Obtain the historical PPG signal database.

[0109] Step 804: For each historical PPG signal data, input the historical PPG signal data into the initial identity recognition model to obtain the initial recognition result output by the initial identity recognition model.

[0110] Step 806: Iteratively train the initial identity recognition model according to the preset iteration conditions to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model.

[0111] Step 808: Obtain the adjusted weight matrix based on the recognition accuracy of the historical identity recognition model.

[0112] Optionally, gradient calculation is performed on the parameters of each model in the historical identity model based on the historical recognition accuracy to obtain the gradient of the first parameter; the gradient matrix of the first parameter is transposed to obtain the gradient matrix of the second parameter; the inner product of the gradient of the first parameter and the gradient of the second parameter is calculated to obtain the adjusted weight matrix.

[0113] Step 810: Obtain the current PPG signal database. The current PPG signal database includes multiple current PPG signal data sets and the current user identity corresponding to each current PPG signal data set.

[0114] Step 812: For each current PPG signal data, the historical identity recognition model is used to classify and identify the current PPG signal data to obtain intermediate recognition results.

[0115] Step 814: For each weight data in the adjusted weight matrix, determine whether the weight data is less than a preset threshold.

[0116] Step 816: If the weight data is less than the preset threshold, the model parameters of the historical identity recognition model are adjusted according to the loss function to obtain the intermediate identity recognition model.

[0117] Step 818: If the weight data is greater than or equal to the preset threshold, then keep the model parameters of the initial identity recognition model unchanged to obtain the intermediate identity recognition model.

[0118] Step 820: Iteratively train the intermediate identity recognition model until the preset iteration conditions are met, and output the intermediate identity recognition model corresponding to the preset iteration conditions to obtain the target identity recognition model.

[0119] Step 822: Acquire target PPG signal data.

[0120] Step 824: Based on the target PPG signal data and the target identity recognition model, obtain the target classification and recognition result corresponding to the target PPG signal data.

[0121] Example 1: The historical PPG signal database contains 12 historical PPG signal data. The current PPG signal database is composed of 10 current PPG signal data, which are 2 historical PPG signal data deleted from the historical PPG signal database.

[0122] The iteration condition was set to 20 iterations. The initial identity recognition model was iteratively trained based on the historical PPG signal database to obtain a historical identity recognition model and a historical identity recognition model with an accuracy of 99.86%.

[0123] Based on the current PPG signal database and the adjusted weight matrix, the historical identity recognition model was iteratively trained to obtain a first target identity recognition model with an accuracy of 99.99%.

[0124] Example 2: The historical PPG signal database includes 12 historical PPG signal data. The current PPG signal database is based on the historical PPG signal database with the addition of 3 historical PPG signal data, which means it includes 15 current PPG signal data.

[0125] The iteration condition was set to 20 iterations. The initial identity recognition model was iteratively trained based on the historical PPG signal database to obtain a historical identity recognition model and a historical identity recognition model with an accuracy of 99.86%.

[0126] Based on the current PPG signal database and the adjusted weight matrix, the historical identity recognition model was iteratively trained to obtain a second target identity recognition model with an accuracy of 97.73%.

[0127] The training results of the historical identity recognition model in Example 1, the first target identity recognition model, and the second target identity recognition model in Example 2 are shown in Table 1.

[0128] Table 1. Training results data for the historical identity recognition model, the first target identity recognition model, and the second target identity recognition model.

[0129] Historical identity recognition model 97.4％ 100 4750 seconds First target identity recognition model 98.1％ 50 540 seconds Second target identity recognition model 95.1％ 50 700 seconds

[0130] As can be seen from Table 1, the training speed of the target identity recognition model trained using the historical identity recognition model is significantly improved, while the accuracy is not significantly reduced and still maintains high recognition performance.

[0131] Using the first and second target identity recognition models, and leveraging the model parameters derived from the historical identity recognition model, allows for faster convergence, a smoother training curve, and less accuracy fluctuation. After 20 iterations, both models have reached a stable state, while the historical model has not yet converged. In terms of training accuracy, the first target identity recognition model achieves a maximum accuracy of 99.99%, superior to the historical model's maximum accuracy of 99.86%. Due to the migration from the smaller historical model to the larger second target identity recognition model, the second model achieves a maximum accuracy of 97.73%, slightly lower than the historical model's accuracy, but without a significant decrease in overall accuracy.

[0132] Additionally, a validation dataset is set up to measure the validation accuracy of the historical identity recognition model in Example 1, the first target identity recognition model, and the second target identity recognition model in Example 2.

[0133] The historical identity recognition model exhibited some fluctuations in verification accuracy at the beginning of training; while the first target identity recognition model and the second target identity recognition model had relatively stable verification accuracy and maintained high accuracy, resulting in improved recognition performance.

[0134] In this embodiment, experiments were conducted on historical identity recognition models and target identity recognition models. The actual experimental results show that the method of this embodiment can improve the training speed of the model and maintain high recognition accuracy when the PPG signal database changes.

[0135] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0136] Based on the same inventive concept, this application also provides an opto-pulse wave-based identity recognition device for implementing the aforementioned opto-pulse wave-based identity recognition method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more opto-pulse wave-based identity recognition device embodiments provided below can be found in the limitations of the opto-pulse wave-based identity recognition method described above, and will not be repeated here.

[0137] In one embodiment, such as Figure 9 As shown, an identity recognition device based on photoplethysmography (PPG) is provided, comprising: a data acquisition module 902 and a classification and recognition module 904, wherein:

[0138] The data acquisition module 902 is used to acquire target PPG signal data.

[0139] The classification and recognition module 904 is used to obtain the target classification and recognition result corresponding to the target PPG signal data based on the target PPG signal data and the target identity recognition model. The target identity recognition model is obtained by training the historical identity recognition model based on the current PPG signal database. The target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted in the process of training the target identity recognition model based on the historical identity recognition model.

[0140] In one embodiment, the apparatus further includes a model training module for iteratively training an initial identity recognition model based on a historical PPG signal database to obtain a historical identity recognition model and the recognition accuracy of the historical identity recognition model; and obtaining an adjustment weight matrix based on the recognition accuracy of the historical identity recognition model.

[0141] In one embodiment, the model training module is further used to acquire a historical PPG signal database, which includes multiple historical PPG signal data and the historical user identities corresponding to each historical PPG signal data; for each historical PPG signal data, the historical PPG signal data is input into an initial identity recognition model to obtain the initial recognition result output by the initial identity recognition model; the initial identity recognition model is iteratively trained according to preset iteration conditions to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model.

[0142] In one embodiment, the model training module is further configured to perform gradient calculation on each model parameter in the historical identity model according to the recognition accuracy to obtain a first parameter gradient matrix; transpose the first parameter gradient matrix to obtain a second parameter gradient matrix; and perform inner product calculation on the first parameter gradient and the second parameter gradient to obtain an adjustment weight matrix.

[0143] In one embodiment, the model training module is further configured to acquire a current PPG signal database and a historical identity recognition model; the current PPG signal database includes multiple target PPG signal data and the target user identity corresponding to each target PPG signal data; for each target PPG signal data, the historical identity recognition model is used to classify and identify the target PPG signal data to obtain intermediate recognition results; based on the intermediate recognition results, the target user identity corresponding to the target PPG signal data, and the loss function, the model parameters of the historical identity recognition model are adjusted to obtain an intermediate identity recognition model; the regularization term of the loss function is determined based on the adjusted weight matrix and the model parameters of the historical identity recognition model; the intermediate identity recognition model is iteratively trained until a preset iteration condition is reached, and the intermediate identity recognition model corresponding to the preset iteration condition is output to obtain the target identity recognition model.

[0144] In one embodiment, the model training module is further configured to, for each weight data in the weight matrix, adjust the model parameters corresponding to the weight data in the historical identity recognition model according to the loss function if the weight data is less than a preset threshold, to obtain an intermediate identity recognition model; if the weight data is greater than or equal to the preset threshold, keep the model parameters corresponding to the weight data in the historical identity recognition model unchanged, to obtain an intermediate identity recognition model; wherein the weight data corresponds to the model parameters of the historical identity recognition model.

[0145] The modules in the aforementioned photoplethysmography (PPG)-based identity recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0146] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements an identification method based on photoplethysmography (PPG). The display screen can be an LCD screen or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0147] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0148] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0149] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0150] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0151] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0152] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0153] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0154] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for identity recognition based on photoplethysmography (PPG), characterized in that, The method includes: Acquire target PPG signal data; Based on the target PPG signal data and the target identity recognition model, the target classification and recognition result corresponding to the target PPG signal data is obtained. The target identity recognition model is obtained by training a historical identity recognition model based on the current PPG signal database. The target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted during the training of the target identity recognition model based on the historical identity recognition model. The historical identity recognition model includes a feature extraction layer and a decision layer. When the scale of the historical PPG signal database and the current PPG signal database are different, the decision layer parameters of the historical identity recognition model are adjusted first, and then the historical identity recognition model is trained based on the current PPG signal database. The historical PPG signal data and the current PPG signal data are PPG signal data with similar edge features and geometric features obtained by preprocessing the raw PPG signals directly collected from the user in a unified manner. The method further includes: The initial identity recognition model is iteratively trained based on the historical PPG signal database to obtain the historical identity recognition model and its recognition accuracy. The adjusted weight matrix is ​​obtained based on the recognition accuracy of the historical identity recognition model; The step of obtaining the adjusted weight matrix based on the recognition accuracy of the historical identity recognition model includes: Based on the recognition accuracy, gradient calculation is performed on the parameters of each model in the historical identity recognition model to obtain the first parameter gradient matrix; The first parameter gradient matrix is ​​transposed to obtain the second parameter gradient matrix; The inner product of the gradients of the first parameter and the second parameter is calculated to obtain the adjusted weight matrix.

2. The method according to claim 1, characterized in that, The step of iteratively training the initial identity recognition model based on the historical PPG signal database to obtain the historical identity recognition model and its recognition accuracy includes: Obtain the historical PPG signal database, which includes multiple historical PPG signal data and the historical user identity corresponding to each historical PPG signal data; For each historical PPG signal data, the historical PPG signal data is input into the initial identity recognition model to obtain the initial recognition result output by the initial identity recognition model; The initial identity recognition model is iteratively trained according to preset iteration conditions to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model.

3. The method according to claim 1, characterized in that, The method further includes: Obtain the current PPG signal database and the historical identity recognition model; the current PPG signal database includes multiple current PPG signal data and the current user identity corresponding to each current PPG signal data. For each current PPG signal data, the historical identity recognition model is used to classify and identify the current PPG signal data to obtain intermediate recognition results; Based on the intermediate identification results, the current user identity corresponding to the current PPG signal data, and the loss function, the model parameters of the historical identity recognition model are adjusted to obtain an intermediate identity recognition model; the regularization term of the loss function is determined based on the adjusted weight matrix and the model parameters of the historical identity recognition model. The intermediate identity recognition model is iteratively trained until a preset iteration condition is reached, and then the intermediate identity recognition model corresponding to the preset iteration condition is output to obtain the target identity recognition model.

4. The method according to claim 3, characterized in that, The step of adjusting the parameters of the historical identity recognition model based on the intermediate recognition result, the current user identity corresponding to the current PPG signal data, and the loss function to obtain the intermediate identity recognition model includes: For each weight data in the adjusted weight matrix, if the weight data is less than a preset threshold, the model parameters corresponding to the weight data in the historical identity recognition model are adjusted according to the loss function to obtain the intermediate identity recognition model. If the weight data is greater than or equal to the preset threshold, then the model parameters corresponding to the weight data in the historical identity recognition model remain unchanged, and the intermediate identity recognition model is obtained. The weight data corresponds to the model parameters of the historical identity recognition model.

5. An identity recognition device based on photoplethysmography (PPG), characterized in that, The device includes: The data acquisition module is used to acquire target PPG signal data; The classification and recognition module is used to obtain the target classification and recognition result corresponding to the target PPG signal data based on the target PPG signal data and the target identity recognition model. The target identity recognition model is obtained by training a historical identity recognition model based on the current PPG signal database. The target identity recognition model includes an adjustment weight matrix, which is used to determine the model parameters that need to be adjusted during the training of the target identity recognition model based on the historical identity recognition model. The historical identity recognition model includes a feature extraction layer and a decision layer. When the scale of the historical PPG signal database and the current PPG signal database are different, the decision layer parameters of the historical identity recognition model are adjusted first, and then the historical identity recognition model is trained based on the current PPG signal database. The historical PPG signal data and the current PPG signal data are PPG signal data with similar edge features and geometric features obtained by preprocessing the raw PPG signals directly collected from the user in a unified manner. The device further includes a model training module, which is used to iteratively train an initial identity recognition model based on a historical PPG signal database to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model; and to obtain the adjusted weight matrix based on the recognition accuracy of the historical identity recognition model. The model training module is further configured to perform gradient calculation on each model parameter in the historical identity recognition model according to the recognition accuracy to obtain a first parameter gradient matrix; transpose the first parameter gradient matrix to obtain a second parameter gradient matrix; and perform inner product calculation on the first parameter gradient and the second parameter gradient to obtain the adjusted weight matrix.

6. The apparatus according to claim 5, characterized in that, The model training module is also used to acquire the historical PPG signal database, which includes multiple historical PPG signal data and the historical user identity corresponding to each historical PPG signal data; for each historical PPG signal data, the historical PPG signal data is input into the initial identity recognition model to obtain the initial recognition result output by the initial identity recognition model; The initial identity recognition model is iteratively trained according to preset iteration conditions to obtain the historical identity recognition model and the recognition accuracy of the historical identity recognition model.

7. The apparatus according to claim 5, characterized in that, The model training module is also used to obtain the current PPG signal database and the historical identity recognition model; the current PPG signal database includes multiple current PPG signal data and the current user identity corresponding to each current PPG signal data; For each current PPG signal data, the historical identity recognition model is used to classify and identify the current PPG signal data to obtain intermediate recognition results. Based on the intermediate recognition results, the current user identity corresponding to the current PPG signal data, and the loss function, the model parameters of the historical identity recognition model are adjusted to obtain an intermediate identity recognition model. The regularization term of the loss function is determined based on the adjusted weight matrix and the model parameters of the historical identity recognition model. The intermediate identity recognition model is iteratively trained until a preset iteration condition is reached, and the intermediate identity recognition model corresponding to the preset iteration condition is output to obtain the target identity recognition model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images