A lightweight mobile PPG signal identification method and system

Abstract

This invention, which lies at the intersection of biometrics and mobile computing, provides a lightweight mobile PPG signal identity recognition method and system. Its technical solution is as follows: a lightweight feature extraction network is trained based on PPG heartbeat cycle segments to obtain a trained lightweight feature extraction network. The lightweight feature extraction network construction process includes: enhancing PPG data; directly inputting the enhanced one-dimensional PPG signal into the input layer of a backbone network to output high-level features; performing decoupling analysis on the high-level features output by the backbone network, performing channel-level gating and feature-level gating calculations, and fusing the calculated results to obtain fused semantic features; and performing identity matching on the PPG signal of the user to be detected based on the trained lightweight feature extraction network to obtain an identity matching result. This significantly reduces computing resource consumption while ensuring recognition accuracy.

Description

Technical Field

[0001] The present invention belongs to the intersection of biometric identification and mobile computing, and in particular relates to a lightweight mobile PPG signal identity recognition method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the increasing popularity of smart wearable devices, biometric identification technology has become an important research area. Photoplethysmography (PPG), a bioelectrical signal, reflects physiological characteristics such as heartbeat by measuring changes in blood flow on the skin's surface. It is non-invasive, easy to collect, and low-cost. Therefore, PPG signals are widely used in health monitoring, heart rate monitoring, blood oxygen saturation monitoring, and other fields.

[0004] Currently, identity recognition methods based on PPG signals face the following challenges:

[0005] (1) The acquisition of PPG signals is affected by the equipment and environment, resulting in significant differences between different devices. In addition, changes in the measurement environment (such as light intensity and temperature) and differences in individual physiological states (such as exercise and emotions) may cause significant differences in the waveform, amplitude, frequency, etc. of PPG signals in different acquisition environments and individual physiological states.

[0006] (2) Traditional PPG signal processing and identification methods mainly rely on complex feature extraction and classification algorithms. These methods usually require high computing resources and energy consumption, and are not suitable for application scenarios with limited computing resources, such as mobile devices. With the rapid development of deep learning technology, neural networks have performed well in pattern recognition and classification tasks. However, deep neural networks are usually complex in structure and have a large number of parameters, which is not conducive to real-time processing on mobile devices. Summary of the Invention

[0007] In order to solve at least one technical problem existing in the above background technology, the present invention provides a lightweight mobile PPG signal identity recognition method and system, which can significantly reduce the consumption of computing resources while ensuring recognition accuracy.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A first aspect of the present invention provides a lightweight mobile PPG signal identity identification method, comprising the following steps:

[0010] Get the user's PPG signal;

[0011] Remove motion artifact components from the preprocessed PPG signal;

[0012] Segment the PPG signal after removing motion artifact components and extract the PPG heart cycle segments;

[0013] The constructed lightweight feature extraction network is trained based on the PPG heartbeat cycle segments to obtain a trained lightweight feature extraction network. The construction process of the lightweight feature extraction network includes:

[0014] Enhance PPG data by adding Gaussian noise to PPG heart cycle segments;

[0015] The data-enhanced one-dimensional PPG signal is directly input into the input layer of the backbone network, and multiple convolution modules are stacked to output high-level features. The high-level features output by the backbone network are decoupled and analyzed, and channel-level gating and feature-level gating are calculated respectively. The calculated results are fused to obtain the fused semantic features.

[0016] Based on the trained lightweight feature extraction network, identity matching is performed on the PPG signal of the user to be detected to obtain the identity matching result.

[0017] Furthermore, removing motion artifact components from the preprocessed PPG signal includes:

[0018] Construct a mixed signal matrix based on the preprocessed PPG signal;

[0019] Perform SDV decomposition on the mixed signal matrix, reconstruct the decomposed signal, and obtain a reconstructed mixed signal matrix;

[0020] The reconstructed mixed signal matrix is ​​converted back to a time domain signal to obtain a reconstructed signal after removing motion artifacts.

[0021] Furthermore, the preprocessing process of the PPG signal includes:

[0022] Perform bandpass filtering on the PPG signal;

[0023] Perform sliding average filtering on the PPG signal after bandpass filtering;

[0024] The PPG signal after sliding average filtering is subjected to power supply interference elimination.

[0025] Furthermore, segmentation is performed based on the PPG signal with motion artifact components removed to extract PPG heart cycle segments, including:

[0026] Set the length of sliding windows and overlapping windows;

[0027] Slide the PPG signal with a set overlapping window. After each sliding step, record the PPG signal heartbeat cycle segments within the sliding window. All heartbeat cycle segments are segmented PPG signals.

[0028] The heartbeat cycle segments are normalized to obtain PPG heartbeat cycle segments.

[0029] Furthermore, the backbone network stacks multiple convolutional modules, each of which can contain a convolutional layer, a batch normalization layer, and an activation function layer, expressed as:

[0030] ,

[0031] in, represents the activation function, represents the batch normalization operation, represents a one-dimensional convolution operation, is the convolution output, is the weight matrix of the convolution kernel, k is the convolution kernel size, is the offset term, and s is the step size.

[0032] Furthermore, the fusion of channel gating and feature-level gating is:

[0033] ,

[0034] ,

[0035] ,

[0036] ,

[0037] ,

[0038] in, represents element-wise multiplication, is the high-level semantic feature of the PPG signal, yes No. Channels in space Elements, Represents the width of Y, represent The number of channels, Y represents the features extracted by the backbone network, yes The elements, is the channel characteristic, is a spatial feature, yes The elements, represents channel-level gating, is feature-level gating, is the sigmoid activation function, and are trainable parameters.

[0039] Furthermore, when training the lightweight feature extraction network, the minimized joint loss function of collaborative training is:

[0040] ,

[0041] ,

[0042] ,

[0043] ,

[0044] in, 、 and is the weight parameter, is the loss function of the features extracted by the backbone network, is the loss function of the fusion feature of channel gating and feature-level gating, for and Perform similarity constraint loss function, Representative characteristics No. i The predicted probability distribution of samples, Representative characteristics No. i The true probability distribution of samples, Represents the first i The predicted probability distribution of samples, Represents the first i The true probability distribution of samples, N is the number of samples.

[0045] A second aspect of the present invention provides a lightweight mobile PPG signal identification system, comprising:

[0046] A signal acquisition module, which is used to obtain the user's PPG signal;

[0047] An artifact removal module is used to remove motion artifact components in the preprocessed PPG signal;

[0048] a segmentation module for performing segmentation based on the PPG signal with motion artifact components removed, and extracting PPG heart cycle segments;

[0049] A lightweight network training module trains a constructed lightweight feature extraction network based on PPG heart cycle segments to obtain a trained lightweight feature extraction network. The lightweight feature extraction network construction process includes: enhancing PPG data by adding Gaussian noise to the PPG heart cycle segments; directly inputting the enhanced one-dimensional PPG signal into the input layer of the backbone network, stacking multiple convolutional modules, and outputting high-level features; performing decoupling analysis on the high-level features output by the backbone network, performing channel-level gating and feature-level gating calculations, and fusing the calculated results to obtain fused semantic features.

[0050] The identity matching module is used to perform identity matching on the PPG signal of the user to be detected based on the trained lightweight feature extraction network to obtain the identity matching result.

[0051] A third aspect of the present invention provides a computer-readable storage medium.

[0052] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the lightweight mobile PPG signal identification method as described above.

[0053] A fourth aspect of the present invention provides a computer device.

[0054] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the lightweight mobile PPG signal identification method described above are implemented.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. This invention decouples and analyzes the high-level features extracted by the backbone network, performing both channel-level gating and feature-level gating calculations on these features. These gating mechanisms are then dynamically weighted to form a unified fusion gating mechanism for feature selection. This design, through the synergistic effect of the gating mechanisms, effectively reduces computational complexity while maintaining feature characterization capabilities, meeting lightweight requirements. While maintaining recognition accuracy, it significantly reduces computing resource consumption. This design makes the method particularly suitable for resource-limited mobile devices, enabling efficient real-time processing of PPG signals.

[0057] 2. Due to the low computational complexity of lightweight neural networks, the present invention operates on mobile devices with low power consumption, helping to extend the device's battery life. This is crucial for extended use, especially in applications such as wearable devices where battery life is critical.

[0058] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0060] Figure 1 This is a flow chart of a lightweight mobile PPG signal identification method provided by an embodiment of the present invention;

[0061] Figure 2 Schematic diagram of a lightweight feature extraction network structure provided by an embodiment of the present invention;

[0062] Figure 3 : This is the ROC curve of the existing MobileNetV4 network and the method of the present invention on the MARSH database provided by the embodiment of the present invention;

[0063] Figure 4 3 is the ROC curve of the existing MobileNetV4 network and the method of the present invention on the noisy MARSH database provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0064] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0065] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0066] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0067] Due to the limited storage capacity and computing resources of mobile devices, PPG signal identification methods for mobile devices should utilize minimal storage and computing resources. Therefore, this paper proposes a lightweight mobile PPG signal identification method. This method first uses a wearable device to collect and preprocess PPG signals, then uses a lightweight neural network to extract discriminative features of the PPG signals. Finally, the Euclidean distance is used to match the identity of the query individual. Due to the advantages of lightweight neural networks, such as high classification accuracy and low memory consumption, they are particularly suitable for identification in mobile environments.

[0068] Example 1

[0069] like Figure 1 As shown, this embodiment provides a lightweight mobile PPG signal identity identification method, including the following steps:

[0070] Step 1: Obtain the user's PPG signal;

[0071] In this embodiment, the user's PPG signal can be collected by a mobile wearable device with a built-in PPG sensor, and then the PPG signal of the user to be detected collected by the mobile wearable device is transmitted to the mobile terminal;

[0072] Specifically, the type of mobile wearable device can be selected according to actual conditions, as long as the built-in PPG sensor can generate data;

[0073] Specifically, when the PPG signal of the user to be detected collected by the mobile wearable device is transmitted to the mobile terminal, the connection transmission can be performed through a wireless transmission method such as Bluetooth, specifically including:

[0074] Bluetooth pairing: When using for the first time, users need to perform Bluetooth pairing between the wearable device and the mobile terminal. This process includes searching for the device and entering the pairing code to ensure the security of the connection.

[0075] Signal acquisition: The signal acquisition sensor emits green or red light to illuminate the skin and receives reflected or transmitted light signals, indirectly measuring changes in blood volume.

[0076] Signal transmission: After completing the collection and processing of the PPG signal, the identity feature data packet is sent to the mobile terminal using the Bluetooth protocol. The data packet should contain information such as timestamp, identity feature code and signal strength.

[0077] Data synchronization: After receiving data from the wearable device, the mobile terminal needs to verify and synchronize the received data, including confirming the integrity and accuracy of the data;

[0078] Continuous reception: The mobile terminal continuously monitors the Bluetooth transmission from the wearable device. Once a data packet is received, it immediately decrypts it and recovers the original PPG signal and related information.

[0079] Step 2: Preprocessing the acquired PPG signal of the user to be detected to obtain a preprocessed PPG signal;

[0080] The collected PPG signals are easily interfered by many factors such as changes in the external environment and human movement, so the PPG signals must be pre-processed by effective methods before use.

[0081] The specific preprocessing process includes the following steps:

[0082] Step 201: performing bandpass filtering on the PPG signal;

[0083] In this embodiment, a 4th-order Butterworth bandpass filter is used to filter high-frequency and low-frequency noise; the lower cutoff frequency , upper cutoff frequency .

[0084] It should be noted that in other embodiments, other band-pass filters or other methods may be used for filtering, as long as the filtering effect can be achieved.

[0085] Step 202: Perform sliding average filtering on the PPG signal after bandpass filtering.

[0086] In this embodiment, a 5-point moving window (window length 100ms) is used to smooth high-frequency glitches.

[0087] Specifically, a 5-point moving window is used to calculate the average value. The 5 points include the previous point and the 2 points before and after the current point (a total of 5 points). The specific calculation formula is:

[0088] ,

[0089] in, represents the signal before sliding smoothing filtering, represents the signal after sliding smoothing filtering, Indicates the position of the previous signal point, Represents the local index within the sliding window.

[0090] Step 203: Eliminate power supply interference from the PPG signal after performing sliding average filtering;

[0091] In this embodiment, a power frequency notch filter is designed to eliminate 50 / 60 Hz power supply interference in the PPG signal.

[0092] Specifically, an infinite impulse response notch filter with a Q value of 35 is used to calculate the bandwidth of 50 Hz and 60 Hz power supply interference respectively, and the double finite difference method is used to calculate the filter coefficients.

[0093] Step 3: Remove motion artifact components from the preprocessed PPG signal;

[0094] The specific steps include:

[0095] Step 301: construct a mixed signal matrix based on the preprocessed PPG signal;

[0096] Specifically, the signal is transformed into Rearrange into an embedding matrix :

[0097] ,

[0098] Where M=N-L+1, N is the signal The total length of , L is the window length.

[0099] Step 302: Perform SDV decomposition on the mixed signal matrix, reconstruct the decomposed signal, and obtain a reconstructed mixed signal matrix;

[0100] Specifically, the decomposition formula is:

[0101] ,

[0102] in, is an L×L orthogonal matrix containing left singular vectors, is an L×M diagonal matrix containing singular values, and V is an M×M orthogonal matrix containing right singular vectors.

[0103] Specifically, suppose that k The singular values ​​and corresponding singular vectors are used to reconstruct the signal. The reconstruction formula is:

[0104] ,

[0105] in, and Only includes the front k List, For the front k A diagonal matrix consisting of singular values.

[0106] Step 303: reconstruct the matrix Convert back to the time domain signal to obtain the reconstructed signal after removing motion artifacts;

[0107] Specifically, this is achieved by averaging the elements of each column:

[0108] , j=1,2,…M.

[0109] in, To remove the motion artifacts, the reconstructed signal is the i-th element in the j-th sliding window of the reconstructed signal.

[0110] Step 4: Segment the PPG signal based on the motion artifact components and extract the PPG heart cycle segments;

[0111] Since the PPG signal waveform collected by the mobile device is greatly affected by external factors and it is difficult to determine the reference point of the PPG signal, in this embodiment, a PPG signal segmentation method based on a sliding window is used to extract PPG heartbeat cycle segments, which specifically includes the following steps:

[0112] Step 401: Set the length of the sliding window and the overlapping window;

[0113] In this embodiment, the sliding window length is set to 1 second, which is longer than one PPG signal cardiac cycle, and the overlapping window length is 0.4 seconds.

[0114] Step 402: Slide the PPG signal with the set overlapping window. After each sliding step, record the PPG signal heart cycle segments within the sliding window. All heart cycle segments are segmented PPG signals.

[0115] Step 403: normalize the heartbeat cycle segments;

[0116] In order to eliminate the influence of different magnitudes of values ​​in different dimensions, in this embodiment, the min-max normalization method is used to normalize the heartbeat cycle segments so that the values ​​corresponding to the heartbeat cycle segments are within the interval [0, 1].

[0117] Step 5: Train the constructed lightweight feature extraction network based on the PPG heartbeat cycle segments to obtain a trained lightweight feature extraction network;

[0118] The specific steps include:

[0119] Step 501: construct a training set and a test set using the extracted PPG heart cycle segments;

[0120] Step 502: Construct a lightweight feature extraction network. The network diagram is as follows: Figure 2 As shown, the specific steps include:

[0121] Step 5021: Enhance the PPG data by adding Gaussian noise to the PPG heart cycle segment.

[0122] In this embodiment, the heartbeat cycle segment of the PPG signal , M is the signal length, the enhanced signal for: ,in, , is the standard deviation of the noise.

[0123] Step 5022: directly input the data-enhanced one-dimensional PPG signal into the input layer of the backbone network, stack multiple convolution modules, and output high-level features;

[0124] In this embodiment, MobileNetV4 is used as the backbone network feature. MobileNetV4 is modified to change all two-dimensional convolution operations to one-dimensional convolution operations.

[0125] Among them, each convolution module can contain a convolution layer, a batch normalization layer (Batch Normalization) and an activation function layer, which can be expressed as:

[0126] ,

[0127] in, represents the activation function, represents the batch normalization operation, represents a one-dimensional convolution operation, is the convolution output, is the weight matrix of the convolution kernel, k is the convolution kernel size, is the offset term, and s is the step size.

[0128] Step 5023: Decoupling analysis is performed on the high-level features output by the backbone network, channel-level gating and feature-level gating are calculated, and the features obtained by the decoupling analysis are fused with the features output by the backbone network to obtain fused semantic features.

[0129] The feature decomposition module aims to decouple and analyze the high-level features extracted by the backbone network. It first calculates channel-level gating and feature-level gating for these high-level features, then fuses them through dynamic weighting to form a unified fusion gating mechanism for feature selection. This design, through the synergistic effect of the gating mechanisms, effectively reduces computational complexity while maintaining feature representation capabilities, meeting lightweight requirements.

[0130] The specific steps include:

[0131] Step 50231: Calculate channel-level gating:

[0132] ,

[0133] ,

[0134] in, yes No. Channels in space Elements, Represents the width of Y, represent The number of channels, Y represents the features extracted by the backbone network, yes The elements, is the channel characteristic, is the sigmoid activation function, are trainable parameters, represents channel-level gating;

[0135] Step 50232: Calculate feature-level gating:

[0136] ,

[0137] ,

[0138] in, Represents spatial features, is a trainable parameter, is feature-level gating;

[0139] Obviously, calculation The number of parameters required is , and the total number of parameters required to calculate channel-level gating and feature-level gating is .

[0140] Step 50233: Fusion of high-level features Y and channel-level gating and feature-level gating :

[0141] ,

[0142] in, represents element-wise multiplication, is the high-level semantic feature of the PPG signal.

[0143] Step 503: Use the training set to train the constructed lightweight feature extraction network, and use the test set to perform identity matching on the PPG signal;

[0144] The test sample dataset is divided into two parts: template sample set and verification sample set. The template sample set generates registration matching templates, and the verification set is used to test the performance of the proposed method.

[0145] In this embodiment, when the feature extraction network is trained, the minimized joint loss function of the collaborative training is:

[0146] ,

[0147] in, 、 and is the weight parameter;

[0148] Features extracted by the backbone network The loss function for:

[0149] ,

[0150] Where N is the number of samples, Representative characteristics No. i The predicted probability distribution of samples, Representative characteristics No. i The true probability distribution of samples;

[0151] The loss function of the fusion feature S of channel gating and feature-level gating for:

[0152] ,

[0153] Where N is the number of samples, Represents the first i The predicted probability distribution of samples, Represents the first i The true probability distribution of samples;

[0154] In order to maintain the intra-class similarity of the same individual and minimize the influence of common features in PPG signal identification, and Apply similarity constraints:

[0155] ,

[0156] After collaborative training, the trained lightweight feature extraction network is finally obtained.

[0157] The specific steps include:

[0158] Step 5031: Obtain the feature distribution code of each user for the template sample set through the trained lightweight feature extraction network, and use it as the registration matching template set. ;

[0159] Step 5032: Verify the PPG signal sample , obtain feature distribution encoding through the trained lightweight feature extraction network ;

[0160] Step 5033: Calculate feature distribution code Matching template set with registration The Euclidean distance between the validation samples The identity label is obtained by the following formula:

[0161] ,

[0162] in, is the indicator function, which only keeps the first i The coefficients corresponding to the class elements are non-zero, and the rest of the elements are all 0.

[0163] Step 504: Evaluate the recognition result.

[0164] In this embodiment, the false acceptance rate (FAR), rejection rate (FRR) and equivalent error rate (EER) are used to measure the effectiveness of the proposed identity recognition method;

[0165] Among them, the error recognition rate is:

[0166] ,

[0167] Rejection rate:

[0168] ,

[0169] Equal error rate:

[0170] ,

[0171] Among them, NGRA is the total number of intra-class tests, NIRA is the total number of inter-class tests; NFR and NFA are the number of false rejections and false acceptances.

[0172] Step 6: Based on the trained lightweight feature extraction network, the PPG signal of the user to be detected is matched to obtain the identity matching result.

[0173] Experimental verification

[0174] The present invention uses the MARSH database and the noisy MARSH database for verification. Figure 3 The figure shows the ROC curves of the existing MobileNetV4 network and the method of the present invention in the MARSH database. Figure 4Figure 2 shows the ROC curves of the existing MobileNetV4 network and the proposed method on the noisy MARSH database. Experiments show that the proposed method achieves the best recognition efficiency on both datasets. This demonstrates that the proposed feature decomposition module and collaborative training strategy can enhance the model's generalization capabilities.

[0175] In summary, the present invention has the advantages of wide application range and easy integration:

[0176] Wide range of applications: The method of the present invention is not only applicable to common mobile devices such as smartphones and smart watches, but can also be extended to other devices with PPG signal acquisition capabilities, such as medical monitoring equipment and fitness trackers. Its wide applicability provides new technical support for identity recognition in multiple fields.

[0177] Easy Integration: Lightweight neural networks have low computational and storage requirements, making the method easy to integrate into existing mobile device platforms. Developers can quickly deploy the identity recognition system through simple hardware and software upgrades, improving device functionality and security.

[0178] Example 2

[0179] This embodiment provides a lightweight mobile PPG signal identification system, including:

[0180] A signal acquisition module, which is used to obtain the user's PPG signal;

[0181] An artifact removal module is used to remove motion artifact components in the preprocessed PPG signal;

[0182] a segmentation module for performing segmentation based on the PPG signal with motion artifact components removed, and extracting PPG heart cycle segments;

[0183] A lightweight network training module trains a constructed lightweight feature extraction network based on PPG heart cycle segments to obtain a trained lightweight feature extraction network. The lightweight feature extraction network construction process includes: enhancing PPG data by adding Gaussian noise to the PPG heart cycle segments; directly inputting the enhanced one-dimensional PPG signal into the input layer of the backbone network, stacking multiple convolutional modules, and outputting high-level features; performing decoupling analysis on the high-level features output by the backbone network, performing channel-level gating and feature-level gating calculations, and fusing the calculated results to obtain fused semantic features.

[0184] The identity matching module is used to perform identity matching on the PPG signal of the user to be detected based on the trained lightweight feature extraction network to obtain the identity matching result.

[0185] It should be noted that the specific implementation method of a lightweight mobile PPG signal identity recognition system in an embodiment of the present invention is similar to the specific implementation method of a lightweight mobile PPG signal identity recognition method in an embodiment of the present invention. Please refer to the description of the method part for details. In order to reduce redundancy, it will not be repeated here.

[0186] Example 3

[0187] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the lightweight mobile PPG signal identification method as described above are implemented.

[0188] Example 4

[0189] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the lightweight mobile PPG signal identification method described above are implemented.

[0190] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0191] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0192] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1The function specified in one or more boxes.

[0193] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0194] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0195] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A lightweight mobile PPG signal identification method, characterized in that: include: Get the user's PPG signal; Remove motion artifact components from the preprocessed PPG signal; Segment the PPG signal after removing motion artifact components and extract the PPG heart cycle segments; The constructed lightweight feature extraction network is trained based on the PPG heartbeat cycle segments to obtain a trained lightweight feature extraction network. The construction process of the lightweight feature extraction network includes: Enhance PPG data by adding Gaussian noise to PPG heart cycle segments; The data-enhanced one-dimensional PPG signal is directly input into the input layer of the backbone network, and multiple convolution modules are stacked to output high-level features. The high-level features output by the backbone network are decoupled and analyzed, and channel-level gating and feature-level gating are calculated respectively. The calculated results are fused to obtain the fused semantic features. Based on the trained lightweight feature extraction network, the identity matching of the user to be detected is performed on the PPG signal to obtain the identity matching result; The fusion of channel gating and feature-level gating is: ， ， ， ， ， in, represents element-wise multiplication, is the high-level semantic feature of the PPG signal, yes No. Channels in space Elements, Represents the width of Y, represent The number of channels, Y represents the features extracted by the backbone network, yes The elements, is the channel characteristic, is a spatial feature, yes The elements, represents channel-level gating, is feature-level gating, is the sigmoid activation function, and is a trainable parameter; When training the lightweight feature extraction network, the minimized joint loss function of collaborative training is: ， ， ， ， in, 、 and is the weight parameter, is the loss function of the features extracted by the backbone network, is the loss function of the fusion feature of channel gating and feature-level gating, for and Perform similarity constraint loss function, Representative characteristics No. i The predicted probability distribution of samples, Representative characteristics No. i The true probability distribution of samples, Represents the first i The predicted probability distribution of samples, Represents the first i The true probability distribution of samples, N is the number of samples.

2. A lightweight mobile PPG signal identification method according to claim 1, characterized in that: Removal of motion artifact components from the pre-processed PPG signal includes: Construct a mixed signal matrix based on the preprocessed PPG signal; Perform SDV decomposition on the mixed signal matrix, reconstruct the decomposed signal, and obtain a reconstructed mixed signal matrix; The reconstructed mixed signal matrix is ​​converted back to a time domain signal to obtain a reconstructed signal after removing motion artifacts.

3. A lightweight mobile PPG signal identification method according to claim 1, characterized in that: The preprocessing process of PPG signal includes: Perform bandpass filtering on the PPG signal; Perform sliding average filtering on the PPG signal after bandpass filtering; The PPG signal after sliding average filtering is subjected to power supply interference elimination.

4. A lightweight mobile PPG signal identification method according to claim 1, characterized in that: Segment the PPG signal after removing motion artifacts and extract the PPG heart cycle segments, including: Set the length of sliding windows and overlapping windows; Slide the PPG signal with a set overlapping window. After each sliding step, record the PPG signal heartbeat cycle segments within the sliding window. All heartbeat cycle segments are segmented PPG signals. The heartbeat cycle segments are normalized to obtain PPG heartbeat cycle segments.

5. A lightweight mobile PPG signal identification method according to claim 1, characterized in that: The backbone network stacks multiple convolutional modules, each of which can contain a convolutional layer, a batch normalization layer, and an activation function layer, expressed as: ， in, represents the activation function, represents the batch normalization operation, represents a one-dimensional convolution operation, is the convolution output, is the weight matrix of the convolution kernel, k is the convolution kernel size, is the offset term, and s is the step size.

6. A lightweight mobile PPG signal identification system, characterized in that: A lightweight mobile PPG signal identification method according to any one of claims 1 to 5 is used, comprising: A signal acquisition module, which is used to obtain the user's PPG signal; An artifact removal module is used to remove motion artifact components in the preprocessed PPG signal; a segmentation module for performing segmentation based on the PPG signal with motion artifact components removed, and extracting PPG heart cycle segments; A lightweight network training module trains a constructed lightweight feature extraction network based on PPG heart cycle segments to obtain a trained lightweight feature extraction network. The lightweight feature extraction network construction process includes: enhancing PPG data by adding Gaussian noise to the PPG heart cycle segments; directly inputting the enhanced one-dimensional PPG signal into the input layer of the backbone network, stacking multiple convolutional modules, and outputting high-level features; performing decoupling analysis on the high-level features output by the backbone network, performing channel-level gating and feature-level gating calculations, and fusing the calculated results to obtain fused semantic features. The identity matching module is used to perform identity matching on the PPG signal of the user to be detected based on the trained lightweight feature extraction network to obtain the identity matching result.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the lightweight mobile PPG signal identification method according to any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps in the lightweight mobile PPG signal identification method according to any one of claims 1 to 5 are implemented.

Images