Intelligent Wearable Vital Sign Detection System and Its Method

Abstract

This application relates to the field of intelligent detection, and specifically discloses an intelligent wearable vital sign detection system and method thereof. By using a neural network model based on deep learning, the hidden feature distribution information of human vital signs in the human pulse signal is mined, so as to accurately express the human vital sign features and improve the accuracy of human vital sign detection.

Description

Technical Field

[0001] This application relates to the field of intelligent detection, and more specifically, to an intelligent wearable vital sign detection system and method thereof. Background Art

[0002] Currently, a large category of applications of wearable devices is for personal vital sign monitoring. Typical products include children's watches, bracelets, and elderly watches, etc. When the existing wearable monitoring systems perform vital sign monitoring of the human body, they need to monitor a relatively large number of vital sign parameters, which increases the influencing factors of the measurement results, not only increasing the technical difficulty and cost, but also ignoring the correlation between each parameter data and reducing the accuracy of vital sign monitoring.

[0003] Therefore, an optimized intelligent wearable vital sign detection system is desired. Summary of the Invention

[0004] To solve the above technical problems, this application is proposed. Embodiments of this application provide an intelligent wearable vital sign detection system and method thereof, which use a neural network model based on deep learning to mine the implicit feature distribution information of human vital signs in the human pulse signal, so as to accurately express the human vital sign features and improve the accuracy of human vital sign detection.

[0005] According to one aspect of this application, an intelligent wearable vital sign detection system is provided, which includes:

[0006] A pulse signal acquisition module, configured to obtain the human pulse signal within a predetermined time period collected by a wearable device worn on the object to be monitored;

[0007] A frequency domain analysis module, configured to perform frequency domain analysis on the human pulse signal based on Fourier transform to obtain a plurality of frequency domain statistical feature values;

[0008] A frequency domain statistical feature extraction module, configured to arrange the plurality of frequency domain statistical feature values into a frequency domain statistical feature vector and then use a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency domain statistical correlation feature vector;

[0009] A time domain feature extraction module, configured to perform image block processing on the waveform diagram of the human pulse signal and then use a ViT model including an embedding layer to obtain a time domain semantic correlation feature vector;

[0010] A time-frequency feature fusion module, configured to fuse the frequency domain statistical correlation feature vector and the time domain semantic correlation feature vector to obtain a classification feature vector;

[0011] The vital sign detection module is used to obtain a classification result by passing the classified feature vector through a classifier, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

[0012] In the above intelligent wearable vital sign detection system, the frequency-domain statistical feature extraction module is used to: use each layer of the convolutional neural network model using a one-dimensional convolutional kernel to respectively perform the following operations on the input data during the forward pass of the layer: perform convolutional processing on the input data to obtain a convolutional feature map; perform pooling based on a feature matrix on the convolutional feature map to obtain a pooled feature map; and perform non-linear activation on the pooled feature map to obtain an activated feature map; wherein, the output of the last layer of the convolutional neural network using a one-dimensional convolutional kernel is the frequency-domain statistical correlation feature vector, and the input of the first layer of the convolutional neural network using a one-dimensional convolutional kernel is the frequency-domain statistical feature vector.

[0013] In the above intelligent wearable vital sign detection system, the time-domain feature extraction module includes: a blocking unit for performing image blocking processing on the waveform diagram of the human pulse signal to obtain a sequence of waveform diagram blocks of the human pulse signal; an image block embedding unit for inputting the sequence of waveform diagram blocks of the human pulse signal into the image block embedding layer of the ViT model including an image block embedding layer to obtain a sequence of waveform diagram block embedding vectors of the human pulse signal; a context encoding unit for passing the sequence of waveform diagram block embedding vectors of the human pulse signal through the ViT module of the ViT model including an image block embedding layer to obtain a plurality of waveform diagram block context semantic correlation feature vectors of the human pulse signal; and a concatenation unit for concatenating the plurality of waveform diagram block context semantic correlation feature vectors of the human pulse signal to obtain the time-domain semantic correlation feature vector.

[0014] In the above intelligent wearable vital sign detection system, the time-frequency feature fusion module includes: an optimization factor calculation unit for calculating the correlation-probability density distribution affine mapping factors between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain a first correlation-probability density distribution affine mapping factor and a second correlation-probability density distribution affine mapping factor; a weighted optimization unit for using the first correlation-probability density distribution affine mapping factor and the second correlation-probability density distribution affine mapping factor as weights to calculate the position-wise weighted sum between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain the classified feature vector.

[0015] In the above intelligent wearable vital sign detection system, the optimization factor calculation unit is configured to: calculate the association - probability density distribution affine mapping factor between the frequency - domain statistical association feature vector and the time - domain semantic association feature vector according to the following optimization formula to obtain the first association - probability density distribution affine mapping factor and the second association - probability density distribution affine mapping factor; where the formula is:

[0016]

[0017]

[0018] where represents the frequency - domain statistical association feature vector, represents the time - domain semantic association feature vector, is the association matrix obtained by the position - by - position association between the frequency - domain statistical association feature vector and the time - domain semantic association feature vector, and are the mean vector and the position - by - position variance matrix of the Gaussian density map formed by the frequency - domain statistical association feature vector and the time - domain semantic association feature vector, represents matrix multiplication, represents the exponential operation of the matrix. The exponential operation of the matrix means calculating the natural exponential function values with the eigenvalues at each position in the matrix as the exponents, represents the first association - probability density distribution affine mapping factor, represents the second association - probability density distribution affine mapping factor.

[0019] In the above intelligent wearable vital sign detection system, the weighted optimization unit is configured to: calculate the position - by - position weighted sum between the frequency - domain statistical association feature vector and the time - domain semantic association feature vector according to the following formula to obtain the classification feature vector; where the formula is:

[0020]

[0021] where represents the frequency - domain statistical association feature vector, represents the time - domain semantic association feature vector, represents the classification feature vector, represents the first association - probability density distribution affine mapping factor, represents the second association - probability density distribution affine mapping factor, represents position - by - position summation.

[0022] In the above-mentioned intelligent wearable vital sign detection system, the vital sign detection module is used to: process the classification feature vector using the classifier according to the following formula to obtain a classification result, where the formula is: , where to is a weight matrix, to is a bias vector, is a classification feature vector.

[0023] According to another aspect of the present application, there is provided an intelligent wearable vital sign detection method, which includes:

[0024] Obtain the human pulse signal within a predetermined time period collected by a wearable device worn on the object to be monitored;

[0025] Perform frequency-domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency-domain statistical feature values;

[0026] Arrange the plurality of frequency-domain statistical feature values into a frequency-domain statistical feature vector and then use a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency-domain statistical correlation feature vector;

[0027] Perform image block processing on the waveform diagram of the human pulse signal and then use a ViT model including an embedding layer to obtain a time-domain semantic correlation feature vector;

[0028] Fuse the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain a classification feature vector;

[0029] Pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

[0030] According to still another aspect of the present application, there is provided an electronic device, including: a processor; and a memory in which computer program instructions are stored, and when the computer program instructions are run by the processor, the processor executes the intelligent wearable vital sign detection method as described above.

[0031] According to yet another aspect of the present application, there is provided a computer-readable medium, on which computer program instructions are stored, and when the computer program instructions are run by a processor, the processor executes the intelligent wearable vital sign detection method as described above.

[0032] Compared with the prior art, an intelligent wearable vital sign detection system and method provided by the present application employ a neural network model based on deep learning to mine the implicit feature distribution information of human vital signs in the human pulse signal, thereby accurately expressing the human vital sign features and improving the accuracy of human vital sign detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0034] Figure 1 FIG. is a schematic diagram of the scenario of the intelligent wearable vital sign detection system according to the embodiment of the present application;

[0035] Figure 2 FIG. is a block diagram of the intelligent wearable vital sign detection system according to the embodiment of the present application;

[0036] Figure 3 FIG. is a system architecture diagram of the intelligent wearable vital sign detection system according to the embodiment of the present application;

[0037] Figure 4 FIG. is a flowchart of the convolutional neural network encoding in the intelligent wearable vital sign detection system according to the embodiment of the present application;

[0038] Figure 5 FIG. is a block diagram of the time-domain feature extraction module in the intelligent wearable vital sign detection system according to the embodiment of the present application;

[0039] Figure 6 FIG. is a block diagram of the time-frequency feature fusion module in the intelligent wearable vital sign detection system according to the embodiment of the present application;

[0040] Figure 7 FIG. is a flowchart of the intelligent wearable vital sign detection method according to the embodiment of the present application;

[0041] Figure 8 FIG. is a block diagram of the electronic device according to the embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0043] OVERVIEW OF THE APPLICATION

[0044] As described above, when the existing wearable monitoring system monitors the vital signs of the human body, it is necessary to monitor more vital sign parameters, which increases the influencing factors of the measurement results, not only increases the technical difficulty and cost, but also ignores the correlation between each parameter data, reducing the accuracy of vital sign monitoring. Therefore, an optimized intelligent wearable vital sign detection system is expected.

[0045] Correspondingly, pulse diagnosis, as a key diagnostic method in traditional Chinese medicine, has a very long history and unique creativity. With the development of the sensing and monitoring field, quantifying and objectifying pulse diagnosis has become a major goal of traditional Chinese medicine pulse diagnosis. The physiological and pathological information of the human body included in the pulse wave, as the basis for clinical diagnosis and treatment, has always been valued by the medical communities at home and abroad. The comprehensive information presented by the pulse wave in terms of morphology (waveform), intensity (amplitude), rate (wave velocity), and rhythm (period) largely reflects many physiological and pathological characteristics of the human cardiovascular system.

[0046] Based on this, in the technical solution of this application, it is expected to use the human pulse signal obtained by the intelligent wearable device to detect the vital signs of the human body. However, considering that there is a large amount of information in the human pulse signal, and the vital sign information of the human body is relatively weak in the pulse wave, which is small-size characteristic information and is easily interfered by external environmental noise, resulting in a low accuracy of human vital sign detection. Therefore, in this process, the difficulty lies in how to extract the implicit feature distribution information of the human vital signs in the human pulse signal, so as to accurately express the characteristics of the human vital signs and improve the accuracy of human vital sign detection.

[0047] In recent years, deep learning and neural networks have been widely applied in fields such as computer vision, natural language processing, and text signal processing. The development of deep learning and neural networks provides new solutions and ideas for extracting the implicit feature distribution information of the human vital signs in the human pulse signal.

[0048] Specifically, in the technical solution of the present application, first, a human pulse signal within a predetermined time period collected by a wearable device worn on the object to be monitored is obtained. Next, considering the human pulse signal, since the waveform diagram of the human pulse signal is a time-domain signal, although the time-domain signal is more intuitive in terms of the dominance of features in the time correlation, due to the weak human pulse signal, it will be interfered by external noise, resulting in a low accuracy of feature extraction for the human pulse signal, thereby affecting the detection and judgment accuracy of human vital signs. The characteristics of the frequency-domain signal are different from those of the time-domain signal. Converting the human pulse signal into the frequency domain can determine the state of human vital signs through the implicit feature distribution information of the human pulse signal in the frequency domain, but its feature dominance in the human pulse signal is not intuitive, ignoring the time-related features. Therefore, in the technical solution of the present application, the combination of the implicit features of the human pulse signal in the time domain and the frequency domain is used to detect human vital signs.

[0049] Specifically, considering that there is a lot of feature information in the human pulse signal and there is a correlation relationship between these feature information, therefore, when extracting the frequency-domain features of the signal, in order to fully excavate the correlation feature information of the feature distribution of the human pulse signal in the frequency domain to improve the detection accuracy of human vital signs, it is necessary to first extract multiple frequency-domain statistical features from the human pulse signal based on the Fourier transform, that is, perform frequency-domain analysis on the human pulse signal based on the Fourier transform to obtain the multiple frequency-domain statistical feature values.

[0050] Then, the multiple frequency-domain statistical feature values are arranged into a frequency-domain statistical feature vector and then feature mining is performed in a convolutional neural network model using a one-dimensional convolution kernel to extract the correlation feature distribution information between the respective frequency-domain statistical feature values, thereby obtaining a frequency-domain statistical correlation feature vector.

[0051] Furthermore, for the human pulse signal, its manifestation in the time domain is a waveform diagram. Therefore, a convolutional neural network model with excellent performance in implicit feature extraction of images is used to mine the features of the waveform diagram of the human pulse signal. However, due to the inherent limitations of convolutional operations, it is difficult for the pure CNN method to learn explicit global and long-range semantic information interaction. Also, considering that the implicit features regarding human vital signs in the waveform diagram of the human pulse signal are small-scale subtle features and are difficult to capture and extract. In order to improve the expression ability of the implicit small-scale subtle features regarding human vital signs in the waveform diagram of the human pulse signal, and thereby improve the accuracy of human vital sign detection, in the technical solution of this application, the waveform diagram of the human pulse signal is subjected to image block processing and then encoded through a ViT model including an embedding layer to extract the implicit context semantic association feature distribution information regarding human vital signs in the waveform diagram of the human pulse signal, thereby obtaining a time-domain semantic association feature vector. It should be understood that in each image block after the image block processing of the waveform diagram of the human pulse signal, the small-scale implicit features regarding human vital signs are no longer small-scale feature information, which is beneficial for subsequent feature extraction of human vital signs. In particular, here, the embedding layer linearly projects each of the image blocks into a one-dimensional embedding vector through a learnable embedding matrix. The process of embedding is to first arrange the pixel values at all pixel positions in each of the image blocks into a one-dimensional vector, and then use a fully connected layer to perform fully connected encoding on this one-dimensional vector to achieve embedding. Also, here, the ViT model can directly process each of the image blocks through a self-attention mechanism to respectively extract the implicit context semantic association feature information regarding the human vital signs in each of the image blocks.

[0052] Then, fuse the frequency-domain statistical association feature vector and the time-domain semantic association feature vector to fuse the association feature information among the respective frequency-domain statistical feature values of the human pulse signal and the implicit context semantic association feature distribution information regarding human vital signs in the waveform diagram of the human pulse signal, so as to perform an optimized expression of the features regarding human vital signs in the human pulse signal, thereby obtaining a classification feature vector.

[0053] Next, further classify the classified feature vector through a classifier to obtain a classification result indicating whether the vital signs of the object to be monitored are normal. That is, in the technical solution of this application, the labels of the classifier include that the vital signs of the object to be monitored are normal (the first label) and that the vital signs of the object to be monitored are abnormal (the second label). Among them, the classifier determines which classification label the classified feature vector belongs to through the softmax function. It should be noted that the first label p1 and the second label p2 here do not include the concept set by humans. In fact, during the training process, the computer model does not have the concept of "whether the vital signs of the object to be monitored are normal". It only has two classification labels and outputs the probabilities of the output features under these two classification labels, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether the vital signs of the object to be monitored are normal is actually transformed into a binary classification probability distribution that conforms to the natural law through the classification label. Substantially, the physical meaning of the natural probability distribution of the label is used, rather than the language text meaning of "whether the vital signs of the object to be monitored are normal". It should be understood that in the technical solution of this application, the classification label of the classifier is the detection and evaluation label of whether the vital signs of the object to be monitored are normal. Therefore, after obtaining the classification result, the human vital signs can be accurately detected based on the classification result.

[0054] Particularly, in the technical solution of this application, here, when fusing the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain the classified feature vector, if the per-position correlation between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector and the consistency of their overall feature distribution under the class probability density distribution can be improved, the fusion effect of the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector can be improved, so as to improve the accuracy of the classification result obtained by the classified feature vector obtained by fusion through the classifier.

[0055] Therefore, the applicant of this application calculates the frequency-domain statistical correlation feature vector, for example, denoted as and the time-domain semantic correlation feature vector, for example, denoted as The correlation-probability density distribution affine mapping factor, expressed as:

[0056]

[0057]

[0058] Is the frequency-domain statistical correlation feature vector And the time-domain semantic correlation feature vector The correlation matrix obtained by the per-position correlation between, and is the mean vector and the per - position variance matrix of the joint Gaussian density map formed by the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector

[0059] That is, by constructing the correlation feature space between the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector and the class probability density space represented by the Gaussian probability density, the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector and can be respectively mapped into the affine homography subspace in the correlation feature space and the class probability density space to extract the representations that conform to the affine homography in the correlation feature domain and the class probability density domain. By using the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector and weight the two vectors respectively with the correlation - probability density distribution affine mapping factor values the per - position correlation of the feature representations of the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector and the consistency in the overall probability density distribution can be improved, thereby enhancing the fusion effect of the frequency - domain statistical correlation feature vector

[0060] and the time - domain semantic correlation feature vector to improve the accuracy of the classification result obtained by the classifier for the fused classification feature vector. In this way, the expression of the human body vital sign features in the human body pulse signal can be optimized to improve the accuracy of the detection of human body vital signs.

[0060] Based on this, the present application proposes an intelligent wearable vital sign detection system, which includes: a pulse signal acquisition module for obtaining a human pulse signal within a predetermined period collected by a wearable device worn on an object to be monitored; a frequency domain analysis module for performing frequency domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency domain statistical feature values; a frequency domain statistical feature extraction module for arranging the plurality of frequency domain statistical feature values into a frequency domain statistical feature vector and then obtaining a frequency domain statistical correlation feature vector through a convolutional neural network model using a one-dimensional convolutional kernel; a time domain feature extraction module for performing image block processing on the waveform diagram of the human pulse signal and then obtaining a time domain semantic correlation feature vector through a ViT model including an embedding layer; a time-frequency feature fusion module for fusing the frequency domain statistical correlation feature vector and the time domain semantic correlation feature vector to obtain a classification feature vector; and a vital sign detection module for passing the classification feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

[0061] Figure 1 FIG. is a schematic diagram of the scenario of the intelligent wearable vital sign detection system according to an embodiment of the present application. As Figure 1 shown, in this application scenario, a human pulse signal within a predetermined period collected by a wearable device worn on an object to be monitored is obtained through a pulse signal sensor (for example, V as Figure 1 illustrated). Then, the above signal is input into a server (for example, S in Figure 1 ) deployed with an intelligent wearable vital sign detection algorithm, where the server can process the above input signal with the intelligent wearable vital sign detection algorithm to generate a classification result for indicating whether the product to be detected meets the anti-counterfeiting requirements.

[0062] After introducing the basic principle of the present application, various non-limiting embodiments of the present application will be specifically introduced below with reference to the accompanying drawings.

[0063] Exemplary System

[0064] Figure 2 FIG. is a block diagram of the intelligent wearable vital sign detection system according to an embodiment of the present application. As Figure 2 shown, the intelligent wearable vital sign detection system 300 according to an embodiment of the present application includes: a pulse signal acquisition module 310; a frequency domain analysis module 320; a frequency domain statistical feature extraction module 330; a time domain feature extraction module 340; a time-frequency feature fusion module 350; and a vital sign detection module 360.

[0065] Among them, the pulse signal acquisition module 310 is used to obtain the human pulse signal within a predetermined period of time collected by a wearable device worn on the object to be monitored; the frequency domain analysis module 320 is used to perform frequency domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency domain statistical feature values; the frequency domain statistical feature extraction module 330 is used to arrange the plurality of frequency domain statistical feature values into a frequency domain statistical feature vector and then use a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency domain statistical correlation feature vector; the time domain feature extraction module 340 is used to perform image block processing on the waveform diagram of the human pulse signal and then use a ViT model including an embedding layer to obtain a time domain semantic correlation feature vector; the time-frequency feature fusion module 350 is used to fuse the frequency domain statistical correlation feature vector and the time domain semantic correlation feature vector to obtain a classification feature vector; the vital sign detection module 360 is used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

[0066] Figure 3 is the system architecture diagram of the intelligent wearable vital sign detection system according to the embodiment of the present application. As Figure 3 shown, in this network architecture, first, the pulse signal acquisition module 310 obtains the human pulse signal within a predetermined period of time collected by a wearable device worn on the object to be monitored; then, the frequency domain analysis module 320 performs frequency domain analysis based on Fourier transform on the human pulse signal obtained by the pulse signal acquisition module 310 to obtain a plurality of frequency domain statistical feature values; the frequency domain statistical feature extraction module 330 arranges the plurality of frequency domain statistical feature values obtained by the frequency domain analysis module 320 into a frequency domain statistical feature vector and then uses a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency domain statistical correlation feature vector; then, the time domain feature extraction module 340 performs image block processing on the waveform diagram of the human pulse signal obtained by the pulse signal acquisition module 310 and then uses a ViT model including an embedding layer to obtain a time domain semantic correlation feature vector; the time-frequency feature fusion module 350 fuses the frequency domain statistical correlation feature vector obtained by the frequency domain statistical feature extraction module 330 and the time domain semantic correlation feature vector obtained by the time domain feature extraction module 340 to obtain a classification feature vector; furthermore, the vital sign detection module 360 passes the classification feature vector fused by the time-frequency feature fusion module 350 through a classifier to obtain a classification result, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

[0067] Specifically, during the operation of the intelligent wearable vital sign detection system 300, the pulse signal acquisition module 310 is configured to obtain the human pulse signal within a predetermined time period collected by the wearable device worn on the object to be monitored. It should be understood that vital signs can be analyzed and detected through the pulse. In one example, the human physiological and pathological information included in the pulse wave serves as the basis for clinical diagnosis and treatment, and the comprehensive information such as the shape (waveform), intensity (amplitude), rate (wave velocity), and rhythm (period) presented by the pulse wave reflects many physiological and pathological characteristics of the human cardiovascular system to a great extent. Therefore, in the technical solution of this application, the human pulse signal can be obtained through the pulse signal sensor deployed in the intelligent wearable device, and the implicit feature distribution information about the human vital signs can be further mined to accurately express the characteristics of the human vital signs, so as to improve the accuracy of the human vital sign detection.

[0068] Specifically, during the operation of the intelligent wearable vital sign detection system 300, the frequency domain analysis module 320 is configured to perform frequency domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency domain statistical feature values. Considering that for the human pulse signal, since the waveform diagram of the human pulse signal is a time domain signal, although the time domain signal is more intuitive for the dominance of features in time correlation, due to the weakness of the human pulse signal, it will be interfered by external noise, resulting in a low accuracy of feature extraction for the human pulse signal, thereby affecting the detection and judgment accuracy of human vital signs. The characteristics of the frequency domain signal are different from those of the time domain signal. Converting the human pulse signal into the frequency domain can determine the state of the human vital signs through the implicit feature distribution information of the human pulse signal in the frequency domain. Therefore, first, in order to fully mine the associated feature information of the feature distribution of the human pulse signal in the frequency domain to improve the detection accuracy of human vital signs, a plurality of frequency domain statistical features are extracted from the human pulse signal based on Fourier transform, that is, frequency domain analysis based on Fourier transform is performed on the human pulse signal to obtain the plurality of frequency domain statistical feature values.

[0069] Specifically, during the operation of the intelligent wearable vital sign detection system 300, the frequency domain statistical feature extraction module 330 is configured to arrange the multiple frequency domain statistical feature values into a frequency domain statistical feature vector and then use a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency domain statistical correlation feature vector. That is, a convolutional neural network model with excellent performance in feature extraction is used to mine the correlation feature distribution information among the respective frequency domain statistical feature values, thereby obtaining a frequency domain statistical correlation feature vector. In a specific example, the convolutional neural network includes a plurality of neural network layers connected in cascade, and each neural network layer includes a convolutional layer, a pooling layer, and an activation layer. Among them, during the encoding process of the convolutional neural network, each layer of the convolutional neural network performs convolutional processing based on the convolutional kernel on the input data, pooling processing on the convolutional feature map output by the convolutional layer using the pooling layer, and activation processing on the pooling feature map output by the pooling layer using the activation layer during the forward propagation of the layer.

[0070] Figure 4 FIG. is a flowchart of convolutional neural network encoding in the intelligent wearable vital sign detection system according to an embodiment of the present application. As Figure 4 shown, during the encoding process of the convolutional neural network, it includes: using each layer of the convolutional neural network model with a one-dimensional convolutional kernel to respectively perform the following operations on the input data during the forward propagation of the layer: S210, performing convolutional processing on the input data to obtain a convolutional feature map; S220, performing pooling based on the feature matrix on the convolutional feature map to obtain a pooling feature map; and S230, performing non-linear activation on the pooling feature map to obtain an activation feature map; wherein, the output of the last layer of the convolutional neural network with a one-dimensional convolutional kernel is the frequency domain statistical correlation feature vector, and the input of the first layer of the convolutional neural network with a one-dimensional convolutional kernel is the frequency domain statistical feature vector.

[0071] Specifically, during the operation of the intelligent wearable vital sign detection system 300, the time-domain feature extraction module 340 is configured to perform image block processing on the waveform diagram of the human pulse signal and then use a ViT model including an embedding layer to obtain a time-domain semantic association feature vector. It should be understood that for the human pulse signal, its manifestation in the time domain is a waveform diagram. Therefore, a convolutional neural network model with excellent performance in implicit feature extraction of images is used to mine the features of the waveform diagram of the human pulse signal. However, due to the inherent limitations of convolutional operations, it is difficult for a pure CNN method to learn explicit global and long-range semantic information interaction. Moreover, it is also considered that since the implicit features related to human vital signs in the waveform diagram of the human pulse signal are small-scale subtle features, it is difficult to capture and extract them. In order to improve the expression ability of the small-scale subtle features related to human vital signs in the waveform diagram of the human pulse signal, thereby improving the accuracy of human vital sign detection, in the technical solution of this application, the waveform diagram of the human pulse signal is subjected to image block processing and then encoded in a ViT model including an embedding layer to extract the implicit context semantic association feature distribution information related to human vital signs in the waveform diagram of the human pulse signal, so as to obtain a time-domain semantic association feature vector. It should be understood that in each image block after the waveform diagram of the human pulse signal is subjected to image block processing, the small-scale implicit features related to human vital signs are no longer small-scale feature information, which is beneficial to subsequent feature extraction of human vital signs. In particular, here, the embedding layer linearly projects each of the image blocks into a one-dimensional embedding vector through a learnable embedding matrix. The implementation process of embedding is to first arrange the pixel values at all pixel positions in each of the image blocks into a one-dimensional vector, and then use a fully connected layer to perform fully connected encoding on this one-dimensional vector to achieve embedding. Moreover, here, the ViT model can directly process each of the image blocks through a self-attention mechanism, so as to extract the implicit context semantic association feature information related to the human vital signs in each of the image blocks respectively.

[0072] Figure 5 Block diagram of the time-domain feature extraction module in the intelligent wearable vital sign detection system according to an embodiment of the present application. As Figure 5As shown in the figure, the time-domain feature extraction module 340 includes: a block unit 341, configured to perform image block processing on the waveform diagram of the human pulse signal to obtain a sequence of waveform diagram blocks of the human pulse signal; an image block embedding unit 342, configured to input the sequence of waveform diagram blocks of the human pulse signal into the image block embedding layer of the ViT model including the image block embedding layer to obtain a sequence of waveform diagram block embedding vectors of the human pulse signal; a context encoding unit 343, configured to pass the sequence of waveform diagram block embedding vectors of the human pulse signal through the ViT module of the ViT model including the image block embedding layer to obtain a plurality of waveform diagram block context semantic association feature vectors of the human pulse signal; and a concatenation unit 344, configured to concatenate the plurality of waveform diagram block context semantic association feature vectors of the human pulse signal to obtain the time-domain semantic association feature vector.

[0073] Specifically, during the operation of the intelligent wearable vital sign detection system 300, the time-frequency feature fusion module 350 is configured to fuse the frequency-domain statistical association feature vector and the time-domain semantic association feature vector to obtain a classification feature vector. That is, after obtaining the frequency-domain statistical association feature vector and the time-domain semantic association feature vector, the two are further subjected to feature fusion to fuse the association feature information between the respective frequency-domain statistical feature values of the human pulse signal and the implicit context semantic association feature distribution information about the human vital signs in the waveform diagram of the human pulse signal, and to perform an optimized expression of the features about the human vital signs in the human pulse signal. In the technical solution of the present application, here, when fusing the frequency-domain statistical association feature vector and the time-domain semantic association feature vector to obtain the classification feature vector, if the position-by-position correlation between the frequency-domain statistical association feature vector and the time-domain semantic association feature vector and the consistency of their overall feature distribution under the class probability density distribution can be improved, the fusion effect of the frequency-domain statistical association feature vector and the time-domain semantic association feature vector can be improved, so as to improve the accuracy of the classification result obtained by the classifier for the fused classification feature vector. Therefore, the applicant of the present application calculates the frequency-domain statistical association feature vector, denoted as and the time-domain semantic association feature vector, denoted as of the association-probability density distribution affine mapping factor, expressed as:

[0074]

[0075]

[0076] where represents the frequency-domain statistical association feature vector, represents the time-domain semantic association feature vector, is the correlation matrix obtained by the position-by-position correlation between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector, and are the mean vector and the position-by-position variance matrix of the Gaussian density map formed by the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector, represents matrix multiplication, represents the exponential operation of a matrix. The exponential operation of the matrix means calculating the values of the natural exponential function with the eigenvalues at each position in the matrix as the exponents, represents the first correlation-probability density distribution affine mapping factor, represents the second correlation-probability density distribution affine mapping factor. That is, by constructing the correlation feature space between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector and the class probability density space represented by the Gaussian probability density, the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector can be respectively mapped into the affine homography subspace in the correlation feature space and the class probability density space to extract the representations that conform to the affine homography in the correlation feature domain and the class probability density domain. By using the correlation-probability density distribution affine mapping factor values and to weight the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector respectively, the position-by-position correlation of the feature representations of the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector and the consistency in the overall probability density distribution can be improved, thereby improving the fusion effect of the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to improve the accuracy of the classification result obtained by the classifier for the fused classification feature vector. In this way, the expression of the human vital sign features in the human pulse signal can be optimized to improve the accuracy of the detection of human vital signs.

[0077] Figure 6 is the block diagram of the time-frequency feature fusion module in the intelligent wearable vital sign detection system according to the embodiment of the present application. As Figure 6As shown, the time-frequency feature fusion module 350 includes: an optimization factor calculation unit 351, configured to calculate the correlation-probability density distribution affine mapping factors between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain a first correlation-probability density distribution affine mapping factor and a second correlation-probability density distribution affine mapping factor; a weighted optimization unit 352, configured to use the first correlation-probability density distribution affine mapping factor and the second correlation-probability density distribution affine mapping factor as weights to calculate the position-wise weighted sum between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain the classification feature vector. More specifically, the weighted optimization unit 352 includes: calculating the position-wise weighted sum between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector according to the following formula to obtain the classification feature vector; where the formula is:

[0078]

[0079] where represents the frequency-domain statistical correlation feature vector, represents the time-domain semantic correlation feature vector, represents the classification feature vector, represents the first correlation-probability density distribution affine mapping factor, represents the second correlation-probability density distribution affine mapping factor, represents the position-wise summation.

[0080] Specifically, during the operation of the intelligent wearable vital sign detection system 300, the vital sign detection module 360 is configured to obtain a classification result by passing the classification feature vector through a classifier, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal. That is, passing the classification feature vector through a classifier to obtain a classification result for indicating whether the vital signs of the object to be monitored are normal. Specifically, the classifier is used to process the classification feature vector according to the following formula to obtain a classification result, where the formula is:

[0081] , where, to is the weight matrix, to is the bias vector, It is a classification feature vector. Specifically, the classifier includes multiple fully-connected layers and a Softmax layer cascaded with the last fully-connected layer of the multiple fully-connected layers. Among them, in the classification process of the classifier, the multiple fully-connected layers of the classifier are used to perform multiple fully-connected encodings on the classification feature vector to obtain an encoded classification feature vector; furthermore, the encoded classification feature vector is input into the Softmax layer of the classifier, that is, the Softmax classification function is used to perform classification processing on the encoded classification feature vector to obtain a classification label. In the technical solution of this application, the labels of the classifier include that the vital signs of the object to be monitored are normal (the first label), and that the vital signs of the object to be monitored are abnormal (the second label). Among them, the classifier determines which classification label the classification feature vector belongs to through the softmax function. It should be noted that the first label p1 and the second label p2 here do not contain artificially set concepts. In fact, during the training process, the computer model does not have the concept of "whether the vital signs of the object to be monitored are normal". It only has two classification labels and the probabilities of the output features under these two classification labels, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether the vital signs of the object to be monitored are normal is actually transformed into a binary classification probability distribution that conforms to the natural law through the classification label. Essentially, the physical meaning of the natural probability distribution of the label is used, rather than the language text meaning of "whether the vital signs of the object to be monitored are normal". It should be understood that in the technical solution of this application, the classification label of the classifier is the detection and evaluation label of whether the vital signs of the object to be monitored are normal. Therefore, after obtaining the classification result, the human vital signs can be accurately detected based on the classification result.

[0082] In summary, the intelligent wearable vital sign detection system 300 according to the embodiment of this application is clarified. It mines the implicit feature distribution information about human vital signs in the human pulse signal by adopting a neural network model based on deep learning, so as to accurately express the features of the human vital signs and improve the accuracy of detecting human vital signs.

[0083] As described above, the intelligent wearable vital sign detection system according to the embodiment of this application can be implemented in various terminal devices. In one example, the intelligent wearable vital sign detection system 300 according to the embodiment of this application can be integrated into the terminal device as a software module and / or a hardware module. For example, the intelligent wearable vital sign detection system 300 can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the intelligent wearable vital sign detection system 300 can also be one of the many hardware modules of the terminal device.

[0084] Alternatively, in another example, the intelligent wearable vital sign detection system 300 and the terminal device can also be separate devices, and the intelligent wearable vital sign detection system 300 can be connected to the terminal device through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0085] Exemplary method

[0086] Figure 7 is a flowchart of an intelligent wearable vital sign detection method according to an embodiment of the present application. As Figure 7 shown, the intelligent wearable vital sign detection method according to an embodiment of the present application includes the steps of: S110, obtaining a human pulse signal within a predetermined time period collected by a wearable device worn on a subject to be monitored; S120, performing frequency-domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency-domain statistical feature values; S130, arranging the plurality of frequency-domain statistical feature values into a frequency-domain statistical feature vector and then using a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency-domain statistical correlation feature vector; S140, performing image block processing on the waveform diagram of the human pulse signal and then using a ViT model including an embedding layer to obtain a time-domain semantic correlation feature vector; S150, fusing the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain a classification feature vector; S160, passing the classification feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether the vital signs of the subject to be monitored are normal.

[0087] In one example, in the above intelligent wearable vital sign detection method, step S130 includes: using each layer of the convolutional neural network model with a one-dimensional convolutional kernel to respectively perform the following operations on the input data during the forward pass of the layer: performing convolutional processing on the input data to obtain a convolutional feature map; performing pooling based on a feature matrix on the convolutional feature map to obtain a pooled feature map; and performing non-linear activation on the pooled feature map to obtain an activation feature map; where the output of the last layer of the convolutional neural network with a one-dimensional convolutional kernel is the frequency-domain statistical correlation feature vector, and the input of the first layer of the convolutional neural network with a one-dimensional convolutional kernel is the frequency-domain statistical feature vector.

[0088] In one example, in the above-mentioned intelligent wearable vital sign detection method, the step S140 includes: performing image block processing on the waveform diagram of the human pulse signal to obtain a sequence of waveform diagram blocks of the human pulse signal; inputting the sequence of waveform diagram blocks of the human pulse signal into the image block embedding layer of the ViT model including the image block embedding layer to obtain a sequence of waveform diagram block embedding vectors of the human pulse signal; passing the sequence of waveform diagram block embedding vectors of the human pulse signal through the ViT module of the ViT model including the image block embedding layer to obtain a plurality of waveform diagram block context semantic association feature vectors of the human pulse signal; and concatenating the plurality of waveform diagram block context semantic association feature vectors of the human pulse signal to obtain the time-domain semantic association feature vector.

[0089] In one example, in the above-mentioned intelligent wearable vital sign detection method, the step S150 includes: calculating the association-probability density distribution affine mapping factors between the frequency-domain statistical association feature vector and the time-domain semantic association feature vector to obtain a first association-probability density distribution affine mapping factor and a second association-probability density distribution affine mapping factor; using the first association-probability density distribution affine mapping factor and the second association-probability density distribution affine mapping factor as weights to calculate the position-wise weighted sum between the frequency-domain statistical association feature vector and the time-domain semantic association feature vector to obtain the classification feature vector. Among them, calculating the association-probability density distribution affine mapping factors between the frequency-domain statistical association feature vector and the time-domain semantic association feature vector to obtain a first association-probability density distribution affine mapping factor and a second association-probability density distribution affine mapping factor includes: calculating the association-probability density distribution affine mapping factors between the frequency-domain statistical association feature vector and the time-domain semantic association feature vector with the following optimization formula to obtain the first association-probability density distribution affine mapping factor and the second association-probability density distribution affine mapping factor; where the formula is:

[0090]

[0091]

[0092] Where represents the frequency-domain statistical association feature vector, represents the time-domain semantic association feature vector, is the association matrix obtained by the position-wise association between the frequency-domain statistical association feature vector and the time-domain semantic association feature vector, and are the mean vector and the position-wise variance matrix of the Gaussian density map formed by the frequency-domain statistical association feature vector and the time-domain semantic association feature vector, represents matrix multiplication, represents the exponential operation of a matrix, and the exponential operation of the matrix represents calculating the values of the natural exponential function with the eigenvalues at each position in the matrix as the exponents, represents the first correlation - probability density distribution affine mapping factor, represents the second correlation - probability density distribution affine mapping factor; more specifically, using the first correlation - probability density distribution affine mapping factor and the second correlation - probability density distribution affine mapping factor as weights, calculating the position - weighted sum between the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector to obtain the classification feature vector, including: calculating the position - weighted sum between the frequency - domain statistical correlation feature vector and the time - domain semantic correlation feature vector with the following formula to obtain the classification feature vector; where the formula is:

[0093]

[0094] where represents the frequency - domain statistical correlation feature vector, represents the time - domain semantic correlation feature vector, represents the classification feature vector, represents the first correlation - probability density distribution affine mapping factor, represents the second correlation - probability density distribution affine mapping factor, represents position - by - position summation.

[0095] In one example, in the above - mentioned intelligent wearable vital sign detection method, the step S160 includes: using the classifier to process the classification feature vector with the following formula to obtain a classification result, where the formula is: , where, to is the weight matrix, to is the bias vector, is the classification feature vector.

[0096] In summary, the intelligent wearable vital sign detection method according to the embodiments of the present application is clarified. It mines the implicit feature distribution information about human vital signs in the human pulse signal by adopting a neural network model based on deep learning, thereby precisely expressing the features of the human vital signs to improve the accuracy of human vital sign detection.

[0097] Exemplary electronic device

[0098] Next, refer to Figure 8 to describe the electronic device according to the embodiments of the present application.

[0099] Figure 8 The block diagram of an electronic device according to an embodiment of the present application is illustrated.

[0100] As Figure 8 shown, the electronic device 10 includes one or more processors 11 and a memory 12.

[0101] The processor 11 may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0102] The memory 12 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage media, and the processor 11 may run the program instructions to implement the functions in the intelligent wearable vital sign detection system of various embodiments of the present application described above and / or other desired functions. Various contents such as classification feature vectors may also be stored in the computer-readable storage media.

[0103] In one example, the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0104] The input device 13 may include, for example, a keyboard, a mouse, etc.

[0105] The output device 14 may output various information to the outside, including classification results, etc. The output device 14 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0106] Of course, for simplicity, Figure 8 only some of the components in the electronic device 10 related to the present application are shown in , and components such as buses, input / output interfaces, etc. are omitted. In addition, according to specific application scenarios, the electronic device 10 may further include any other appropriate components.

[0107] Exemplary computer program products and computer-readable storage media

[0108] In addition to the above methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions that, when run by a processor, cause the processor to execute the steps in the functions of the intelligent wearable vital sign detection method according to various embodiments of the present application described in the "Exemplary System" section above of this specification.

[0109] The computer program product can be written in any combination of one or more programming languages for the program code to perform the operations of the embodiments of the present application. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0110] In addition, an embodiment of the present application may also be a computer-readable storage medium, on which computer program instructions are stored that, when run by a processor, cause the processor to execute the steps in the functions of the intelligent wearable vital sign detection method according to various embodiments of the present application described in the "Exemplary System" section above of this specification.

[0111] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0112] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present application. In addition, the above-disclosed specific details are only for the purposes of illustration and easy understanding, and are not limitations. The above details do not limit the present application to necessarily adopt the above specific details to implement.

[0113] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc. are open-ended terms that mean "including but not limited to" and can be used interchangeably with each other. The word "or" and "and" used herein refer to the phrase "and / or" and can be used interchangeably with it, unless the context clearly indicates otherwise. The phrase "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with it.

[0114] It should also be noted that in the devices, equipment, and methods of this application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of this application.

[0115] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0116] The above description has been given for purposes of illustration and description. In addition, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although several example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions, and sub-combinations thereof.

Claims

1. An intelligent wearable vital sign detection system, characterized in that, Including: A pulse signal acquisition module, configured to acquire a human pulse signal within a predetermined time period collected by a wearable device worn on an object to be monitored; A frequency domain analysis module, configured to perform frequency domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency domain statistical feature values; A frequency domain statistical feature extraction module, configured to arrange the plurality of frequency domain statistical feature values into a frequency domain statistical feature vector and then use a convolutional neural network model with a one-dimensional convolutional kernel to obtain a frequency domain statistical correlation feature vector; A time domain feature extraction module, configured to perform image block processing on the waveform diagram of the human pulse signal and then use a ViT model including an image patch embedding layer to obtain a time domain semantic correlation feature vector; A time-frequency feature fusion module, configured to fuse the frequency domain statistical correlation feature vector and the time domain semantic correlation feature vector to obtain a classification feature vector; A vital sign detection module, configured to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

2. The intelligent wearable vital sign detection system according to claim 1, characterized in that, The frequency domain statistical feature extraction module is configured to: use each layer of the convolutional neural network model with a one-dimensional convolutional kernel to respectively perform the following operations on the input data during the forward pass of the layer: Perform convolutional processing on the input data to obtain a convolutional feature map; Perform pooling based on a feature matrix on the convolutional feature map to obtain a pooled feature map; And Perform non-linear activation on the pooled feature map to obtain an activated feature map; Wherein, the output of the last layer of the convolutional neural network with a one-dimensional convolutional kernel is the frequency domain statistical correlation feature vector, and the input of the first layer of the convolutional neural network with a one-dimensional convolutional kernel is the frequency domain statistical feature vector.

3. The intelligent wearable vital sign detection system according to claim 2, characterized in that, The time domain feature extraction module includes: A block unit, configured to perform image block processing on the waveform diagram of the human pulse signal to obtain a sequence of waveform diagram blocks of the human pulse signal; An image patch embedding unit, configured to input the sequence of waveform diagram blocks of the human pulse signal into the image patch embedding layer of the ViT model including the image patch embedding layer to obtain a sequence of waveform diagram block embedding vectors of the human pulse signal; A context encoding unit, configured to pass the sequence of waveform diagram block embedding vectors of the human pulse signal through the ViT module of the ViT model including the image patch embedding layer to obtain a plurality of waveform diagram block context semantic correlation feature vectors of the human pulse signal; and A concatenation unit, configured to concatenate the plurality of waveform diagram block context semantic correlation feature vectors of the human pulse signal to obtain the time domain semantic correlation feature vector.

4. The intelligent wearable vital sign detection system according to claim 3, wherein The time-frequency feature fusion module includes: An optimization factor calculation unit, configured to calculate an association-probability density distribution affine mapping factor between the frequency domain statistical correlation feature vector and the time domain semantic correlation feature vector to obtain a first association-probability density distribution affine mapping factor and a second association-probability density distribution affine mapping factor; A weighted optimization unit, which is configured to calculate a position-wise weighted sum between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector by using the first correlation-probability density distribution affine mapping factor and the second correlation-probability density distribution affine mapping factor as weights, so as to obtain the classification feature vector.

5. The intelligent wearable vital sign detection system according to claim 4, wherein, The optimization factor calculation unit is configured to: calculate the correlation-probability density distribution affine mapping factors between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector by using the following optimization formula to obtain the first correlation-probability density distribution affine mapping factor and the second correlation-probability density distribution affine mapping factor; wherein, the formula is: wherein represents the frequency-domain statistical correlation feature vector, represents the time-domain semantic correlation feature vector, is the correlation matrix obtained by position-wise correlation between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector, and are the mean vector and the position-wise variance matrix of the Gaussian density map formed by the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector, represents matrix multiplication, represents the exponential operation of a matrix. The exponential operation of the matrix means calculating the values of the natural exponential function with the eigenvalues at each position in the matrix as the exponents, represents the first correlation-probability density distribution affine mapping factor, represents the second correlation-probability density distribution affine mapping factor.

6. The intelligent wearable vital sign detection system according to claim 5, characterized in that, The weighted optimization unit is configured to: calculate the position-wise weighted sum between the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector by using the following formula to obtain the classification feature vector; wherein, the formula is: wherein represents the frequency-domain statistical correlation feature vector, represents the time-domain semantic correlation feature vector, represents the classification feature vector, represents the first correlation - probability density distribution affine mapping factor, represents the second correlation - probability density distribution affine mapping factor, represents summation by position.

7. The intelligent wearable vital sign detection system according to claim 6, characterized in that, The physical sign detection module is configured to: use the classifier to process the classification feature vector by using the following formula to obtain a classification result, wherein, the formula is: , where to is the weight matrix, to is the bias vector, is the classification feature vector.

8. An intelligent wearable vital sign detection method, characterized in that, including: Obtain a human pulse signal within a predetermined time period collected by a wearable device worn on an object to be monitored; Perform frequency-domain analysis based on Fourier transform on the human pulse signal to obtain a plurality of frequency-domain statistical feature values; Arrange the plurality of frequency-domain statistical feature values into a frequency-domain statistical feature vector, and then obtain a frequency-domain statistical correlation feature vector through a convolutional neural network model using a one-dimensional convolutional kernel; Perform image block processing on the waveform diagram of the human pulse signal, and then obtain a time-domain semantic correlation feature vector through a ViT model including an image patch embedding layer; Fuse the frequency-domain statistical correlation feature vector and the time-domain semantic correlation feature vector to obtain a classification feature vector; Pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal.

9. The intelligent wearable vital sign detection method according to claim 8, characterized in that, Performing image block processing on the waveform diagram of the human pulse signal and then obtaining a time-domain semantic correlation feature vector through a ViT model including an image patch embedding layer includes: Performing image block processing on the waveform diagram of the human pulse signal to obtain a sequence of waveform diagram blocks of the human pulse signal; Input the sequence of waveform diagram blocks of the human pulse signal into the image patch embedding layer of the ViT model including the image patch embedding layer to obtain a sequence of waveform diagram block embedding vectors of the human pulse signal; Pass the sequence of waveform diagram block embedding vectors of the human pulse signal through the ViT module of the ViT model including the image patch embedding layer to obtain a plurality of waveform diagram block context semantic correlation feature vectors of the human pulse signal; and Cascade the plurality of waveform diagram block context semantic correlation feature vectors of the human pulse signal to obtain the time-domain semantic correlation feature vector.

10. The intelligent wearable vital sign detection method according to claim 9, characterized in that, Passing the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the vital signs of the object to be monitored are normal, includes: using the classifier to process the classification feature vector by using the following formula to obtain a classification result, wherein, the formula is: , where to is the weight matrix, to is the bias vector, is the classification feature vector.

Images