Human health monitoring method based on wearable device

Abstract

The invention discloses a human health monitoring method based on wearable equipment. The human health monitoring method comprises the following steps: S1, acquiring physiological data of a target object; s2, adding a health state label to the physiological data; converting the physiological data into a time sequence according to the collection time, and recording the time sequence as a physiological time sequence; s3, performing data enhancement processing on the physiological time sequence, and evaluating the enhanced physiological time sequence by using a Bayesian method and information entropy; s4, evaluating the physiological reconstruction sequence, and determining the reconstruction effect of the physiological reconstruction sequence; and S5, performing feature extraction on the physiological reconstruction sequence, and inputting the extracted features into the bidirectional long-short memory network to obtain a classification result of the physiological reconstruction sequence. The beneficial effects of the invention are that the efficiency of data processing and classification is improved, and the reliability and practicability of health state monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data monitoring, and in particular to a human health monitoring method based on a wearable device. Background Art

[0002] As the problem of population aging becomes increasingly serious, how to maintain and improve the health of the target object has become an important task for society. In this context, smart health wearable devices aimed at health monitoring and disease prevention have emerged. This type of wearable device relies on the collection, processing and analysis of multimodal physiological signals to achieve a comprehensive assessment and monitoring of individual health status. Physiological signals are an important source of information reflecting the individual's physiological condition, including but not limited to electrocardiogram signals, blood pressure signals, blood oxygen saturation signals, body temperature signals, etc. These signals can reflect the working status of various organs and systems in the human body, so they are widely used in the fields of medical and health management.

[0003] However, the analysis and processing of physiological signals collected by wearable devices is an extremely complex task. On the one hand, physiological signals often contain a lot of noise and interference, such as noise from the environment, interference from the equipment, and bioelectric interference from the human body. On the other hand, the pattern of physiological signals is extremely complex and contains rich temporal and spatial information. Traditional signal processing methods alone often cannot extract effective features. Therefore, how to effectively process and analyze physiological signals collected by wearable devices and extract useful information has become an important research topic. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above problems or problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the purpose of the present invention is to provide a human health monitoring method based on wearable devices, which identifies the noise existing in the processing of physiological health data of the target object through signal processing and machine learning, thereby improving the reliability of data processing.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a human health monitoring method based on a wearable device, comprising:

[0008] S1, obtaining physiological data of the target object;

[0009] S2, adding health status labels to physiological data; converting physiological data into time series according to the time of collection, recorded as physiological time series;

[0010] S3, performing data enhancement processing on the physiological time series, using the Bayesian method and information entropy to evaluate the physiological time series, and reconstructing the evaluated physiological time series into a physiological reconstruction sequence;

[0011] S4, evaluating the physiological reconstruction sequence to determine the reconstruction effect of the physiological reconstruction sequence;

[0012] S5, extracting features from the physiological reconstruction sequence, inputting the extracted features into a bidirectional long short-term memory network, and obtaining a classification result of the physiological reconstruction sequence.

[0013] The beneficial effects of the present invention are as follows: the present invention ensures the validity of enhanced data by using the Bayesian method to evaluate the posterior probability and the prior probability; the compressed sensing parameters are calculated by information entropy, and the parameters are dynamically adjusted to optimize the reconstruction effect; data enhancement increases the diversity of samples, improves the generalization ability and robustness of the model, thereby improving the accuracy of processing and classification of the physiological data of the target object, and improving the reliability and practicality of health status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0015] Figure 1 The figure is a flowchart of a human health monitoring method based on wearable devices.

[0016] Figure 2 A logical schematic diagram of feature extraction for a human health monitoring method based on wearable devices.

[0017] Figure 3 Schematic diagram of bidirectional conduction of a bidirectional long short-term memory network for a human health monitoring method based on wearable devices.

[0018] Figure 4 Schematic diagram of step S3 of a human health monitoring method based on a wearable device. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0022] Reference Figure 1 This embodiment provides a human health monitoring method based on a wearable device, including:

[0023] S1, obtaining physiological data of the target object.

[0024] The acquired physiological data include acquired heart rate, blood pressure, and corresponding status identifiers; the status identifier is used to identify the data type corresponding to the acquired physiological data, such as whether it is exercise, sitting still, or sleeping, and the recorded data is heart rate and blood pressure; at the same time, data with large changes in physiological data will be marked in advance to identify and process these data that may be abnormal.

[0025] At this time, specific wearable monitoring devices are used to collect physiological data of the target object. These devices include smart bracelets, smart watches, smart blood pressure monitors, etc., which have sensors and data recording functions and can collect data on various physiological indicators.

[0026] These devices typically include the following features:

[0027] Heart rate monitoring: Real-time monitoring of the target object's heart rate changes through the built-in heart rate sensor or photoelectric sensor.

[0028] Blood pressure monitoring: Some devices can provide blood pressure measurement capabilities, using pulse wave technology or photoelectric sensors to estimate the target subject's blood pressure value.

[0029] Exercise monitoring: The device can record the target subject’s steps, activity duration, calorie consumption and other exercise-related information to assess the target subject’s daily exercise volume, while also recording changes in physiological data during exercise to obtain relevant data during exercise.

[0030] Sleep monitoring: The accelerometer and light sensor can be used to monitor the target subject’s sleep quality, sleep duration, and the proportion of different sleep stages.

[0031] The target subject wears the device on his body, and the device will periodically record and store physiological data. The data format is signal data, which is recorded in the form of time series.

[0032] The time series recorded at this time will be based on the corresponding situation at the time of recording. The acquired heart rate signal will be labeled according to the different acquisition methods each time. At the same time, the corresponding fluctuation form in the current data will be recorded in time to record the changes in the corresponding data when the elderly person's condition changes.

[0033] S2, adding health status labels to physiological data; converting physiological data into time series according to the time of collection, recorded as physiological time series.

[0034] Furthermore, in order to conduct supervised learning, the collected physiological data are added with corresponding health status labels, including normal, sub-health, hypertension, arrhythmia, and physical fatigue.

[0035] Furthermore, the labeled physiological data is converted into a time series according to the time of collection. The time series obtained at this time can be represented by expressing the acquired heart rate and other data in the form of an ECG data signal to obtain the fluctuations corresponding to different data in the physiological data, so as to understand the changing trend of the data within the corresponding time.

[0036] In one embodiment, the physiological time series ECG(t) can be expressed as:

[0037] ECG(t)={V1,V2,V3,…,V n};

[0038] Among them, ECG(t) represents the physiological time series within the range of time t, V1, V2, V3, …, V n Represents the measured values ​​at the corresponding time points 1, 2…n, where n is the length of the physiological time series.

[0039] For example, in this embodiment, the physiological time series collected at a certain moment is ECG(t)={0.5, 0.6, 0.55, ...}. This sequence represents the values ​​collected at the corresponding time series. When the collected values ​​are converted into a time series, all data will be normalized according to the standard value to eliminate the dimension of the data.

[0040] Furthermore, the physiological time series is annotated. The present invention uses a binary classification method to label the data to distinguish normal data from abnormal data. The annotation can be expressed as:

[0041]

[0042] Among them, Label(ECG(t)) represents the label of the physiological time series, which is used to mark all recognized signals to identify the corresponding data.

[0043] S3, performs data enhancement processing on the physiological time series, uses the Bayesian method and information entropy to evaluate the physiological time series, and reconstructs the evaluated physiological time series into a physiological reconstruction sequence.

[0044] like Figure 4 As shown, in one embodiment of the present invention, the processing method of data enhancement processing in step S3 includes:

[0045] S31, perform wavelet transformation on the physiological time series, and use the Bayesian function to verify the posterior probability and prior probability of the physiological time series.

[0046] The processing method for data enhancement of physiological time series is to use an improved compressed sensing algorithm for data enhancement. Specifically, firstly, the transform domain of the physiological time series is sparsely represented and converted into the format required for processing. Then, the sparsely represented physiological time series can be expressed as:

[0047] S=arg min∥W(ECG(t))∥0;

[0048] Among them, ∥·∥0 represents the L0 norm, which is used to measure the number of non-zero elements in the vector. W() represents the wavelet transform, which transforms the physiological time series into the required sparse form; S is the sparsely represented physiological time series, so that the non-zero components of the signal are the least in the selected transform domain; W(ECG(t)) represents the physiological time series after wavelet transform.

[0049] Furthermore, the Bayesian method and information entropy are used to evaluate the uncertainty of the data, and the key parameters of the compressed sensing algorithm are adjusted accordingly. At this time, the posterior probability in the Bayesian method is used to verify the classification results of the physiological time series at this time. The posterior probability is based on the data corresponding to the prior probability and the likelihood function for verification. The purpose is to determine whether the physiological time series obtained at this time can conform to the rules of normal processing, so as to identify whether there is abnormal data in it. The posterior probability corresponding to the physiological time series can be expressed as:

[0050]

[0051] Among them, p(S|D) represents the posterior probability corresponding to the physiological time series, p(D|S) represents the likelihood function corresponding to the physiological time series, p(S) represents the prior probability corresponding to the physiological time series, and p(D) represents the marginal probability, which is a normalization constant used to ensure that the sum of the posterior probability after calculation is 1.

[0052] The calculation method of the prior probability is realized through the sparse representation of the physiological time series, and the prior probability is expressed as:

[0053]

[0054] Among them, λ represents the sparsity parameter, which is used to control the sparsity of the current physiological time series, e represents the exponential constant, S i Represents the value of the element in the i-th sparsely represented physiological time series.

[0055] The obtained likelihood function can be expressed as:

[0056]

[0057] Among them, σ represents the standard deviation of the measured value of the physiological time series, y i represents the i-th measurement value in the physiological time series, ψ represents the sparse change matrix, which is used for sparse change, and the size of the sparse change matrix is ​​the same as the size of the physiological time series; φ represents the measurement matrix, and the size of the measurement matrix is ​​the same as the size of the physiological time series; S represents the sparsely represented physiological time series; (ψφS) i represents the product of the corresponding values ​​in the i-th measurement matrix, the sparse change matrix and the sparsely represented physiological time series, and e represents the exponential constant.

[0058] S32, obtaining the information entropy of the physiological time series according to the calculated prior probability.

[0059] The way to obtain information entropy can be expressed as:

[0060] H(S)=-∑p(S)logp(S);

[0061] Among them, H(S) is the information entropy, which is used to evaluate the uncertainty of S.

[0062] S33, calculating compressed sensing parameters according to the acquired information entropy, and reconstructing the physiological time series using the compressed sensing parameters to obtain a physiological reconstructed sequence.

[0063] The adjustment strategy of the compressed sensing parameters of the compressed sensing algorithm can be expressed as:

[0064]

[0065] Among them, θ new represents the adjusted compressed sensing parameters, θ old represents the compressed sensing parameters before adjustment, α is the learning rate, which controls the speed of parameter update. By evaluating the uncertainty of S, the compressed sensing parameters are dynamically adjusted; Represents the gradient operator of the compressed sensing function.

[0066] Furthermore, compressed sensing parameters are applied to reconstruct the data and reconstruct and enhance the physiological time series.

[0067] The physiological time series is reconstructed as follows:

[0068] r k =y k -S k θ k ;

[0069] Among them, r k represents the residual of the physiological time series after the kth iteration, θ k represents the compressed sensing parameters after the kth iteration; y k represents the measured value after the kth iteration in the physiological time series; S k Represents the value of the element in the sparsely represented physiological time series of the kth iteration; at this time, the value of the corresponding residual in the physiological time series is continuously verified, and when the corresponding compressed sensing parameters are known, the value of the physiological time series is continuously adjusted until the calculated residual meets the current calculation requirements.

[0070] When verifying the residuals, if the residuals do not meet the requirements, it is necessary to iteratively change the values ​​in the physiological time series. The change can be:

[0071]

[0072] Among them, y k It represents the adjustment value after the kth iteration in the physiological time series. This adjustment value is adjusted when the calculated residual does not meet the preset value. At this time, the calculated value is the value in the corresponding physiological reconstruction sequence; S i,k represents the value of the sparse representation of the i-th element in the physiological time series at the k-th iteration, represents the average value in the sparsely represented physiological time series of the kth iteration, θ i,k represents the compressed sensing parameter corresponding to the kth iteration of the ith element in the physiological time series, represents the average value of the compressed sensing parameters after the kth iteration, represents the adjustment coefficient at the kth iteration, and is used to reconstruct the values ​​in the physiological time series so as to obtain the corresponding physiological reconstruction sequence. K represents the number of iterations, n represents the number of elements in the physiological time series, and k=1,2,,K.

[0073] S4, evaluating the physiological reconstruction sequence to determine the reconstruction effect of the physiological reconstruction sequence.

[0074] Furthermore, the quality of the enhanced data is evaluated, such as by comparing the difference between the original data and the reconstructed data to verify the reconstruction effect.

[0075] The quality of the enhanced data is assessed by using mean square error to quantify the differences in the data.

[0076] At this time, the method of using mean square error for adjustment is:

[0077]

[0078] Among them, MSE represents mean square error, represents the value of the jth physiological reconstruction sequence, m represents the number of data in the physiological reconstruction sequence, represents the average value of the physiological reconstruction sequence, and the value of j ranges from 1 to m.

[0079] Verify whether the mean square error at this time is the same as the expected value, so as to obtain the physiological reconstruction sequence with the minimum mean square error as the physiological reconstruction sequence output at this time.

[0080] S5, extracting features from the physiological reconstruction sequence, inputting the extracted features into a bidirectional long short-term memory network, and obtaining a classification result of the physiological reconstruction sequence.

[0081] like Figure 2 As shown, the data after data enhancement is set to array X = {x1, x2, ..., x j ,…,x m}, and input the array into the feature extraction network model for feature extraction. The feature extraction neural network model includes an input layer, a bidirectional long short-term memory network layer, a fully connected layer, a convolutional layer, an attention enhancement layer, and an output layer. The network model is used to extract features from the physiological reconstruction sequence.

[0082] like Figure 3 As shown in the figure, the physiological reconstruction sequence is first input, and the temporal features of the physiological feature sequence are selectively screened through bidirectional propagation through the bidirectional long short-term memory network. The bidirectional long short-term memory network has the structural characteristics of forward propagation and backward propagation. The feature recognition of the physiological reconstruction sequence is performed on the basis of the bidirectional structure. The processing method at this time can obtain features from two directions while ensuring the stability of the network, thereby improving the efficiency of feature screening.

[0083] The bidirectional long short-term memory network mainly extracts the time domain correlation in the physiological reconstruction sequence. Its core is to use memory blocks to memorize the input data for a long or short period of time, and manage it through a gate mechanism to achieve a multi-level feature selection. Specifically, the bidirectional long short-term memory network layer uses c(t) as the long memory time axis and h(t) as the short memory time axis. It selectively forgets the input data through the long short-term memory function, thereby retaining effective features and eliminating invalid or repeated information. If the hidden state of the forward propagation output of the bidirectional long short-term memory network at time t is a, and the reverse output hidden state is b, then the output of the bidirectional long short-term memory network is the comprehensive output of the forward network and the backward network at each moment, and the hidden state of the total output can be expressed as:

[0084]

[0085] in, is a fully connected operation; h t is the total output hidden state at time point t, indicating the network's encoding of the input signal at that moment; h a is the hidden state output of the forward LSTM network at the same time point t; h b It is the hidden state output of the backward long short-term memory network at time point t; the total output hidden state output at this time is the temporal relationship of the output physiological reconstruction sequence, and the output value at this time can reflect the corresponding temporal situation.

[0086] Furthermore, the output of the bidirectional long short-term memory network is passed into the fully connected layer and a new time-frequency matrix is ​​obtained by time-frequency transformation. When the time-frequency matrix passes through a convolutional layer composed of multiple convolutional layers and maximum pooling layers stacked alternately, the feature extraction of the physiological reconstruction sequence is completed.

[0087] Furthermore, the extracted temporal features are input into the attention enhancement layer, and then classified using the Softmax classification function to obtain the classification probability of the physiological reconstruction sequence, and the corresponding classification result is determined according to the obtained classification probability.

[0088] Furthermore, the convolution layers at this time are: the first layer: 32 convolution kernels, size 3x3; the second layer: 64 convolution kernels, size 3x3; the third layer: 128 convolution kernels, size 3x3; each convolution layer is followed by a maximum pooling layer, and the pooling window size is 2x2.

[0089] When the Softmax classification function performs classification, the ratio of the exponential value of the output vector of the fully connected layer to the sum of the exponential values ​​is used as the classification probability recognized at this time, so that the probability distribution of the corresponding data can be obtained.

[0090] At this time, the final classification probability is expressed as:

[0091]

[0092] Among them, CA represents the final classification probability, which is a vector of length CC, and each element represents the probability of the corresponding category; Softmax represents the Softmax classification function, which is used to convert the output of the fully connected layer into a probability distribution; z b Represents the b-th element in the fully connected layer output vector z, b = 1, 2,, C; C represents the number of categories, that is, the number of different categories in the dataset.

[0093] The classification probability output at this time can indicate the condition of the physiological data of the target object currently being detected, so that timely warning can be given according to the condition of the target object, thereby taking corresponding emergency measures.

[0094] After obtaining the classification probability, the classification result is obtained by successively judging the change value of the numerical value calculated by the classification probability, and determining the correlation between the classification probability of the physiological reconstruction sequence and the currently input physiological time series within a preset time period as the first correlation; based on the obtained first correlation, extracting the second correlation from the first correlation; and obtaining the output classification result based on the obtained first correlation and the second correlation.

[0095] The implementation method for calculating the correlation between the classification probability of the physiological reconstruction sequence and the currently input physiological time series is to sort the physiological time series according to the classification probability, and calculate the Pearson correlation coefficient between the physiological reconstruction sequence and the physiological time series under each classification probability in turn as the first correlation. At the same time, in order to improve the calculation effect of the first correlation, the residual of the physiological time series and the physiological reconstruction sequence is used to calculate the second correlation, and the classification corresponding to the maximum value between the first correlation and the second correlation is used as the output classification result.

[0096] The residual here is the corresponding difference obtained by sequentially calculating the physiological time series and the physiological reconstruction series with the corresponding preset values.

[0097] At this time, the first correlation is calculated through the Pearson correlation coefficient. The values ​​in the physiological time series and the physiological reconstruction sequence are used as input, and the predicted ideal value is used as the basis for comparison at this time to calculate the Pearson correlation coefficient. The Pearson correlation coefficient is a prior art and will not be demonstrated here.

[0098] The second correlation at this time is calculated by calculating the correlation between the residual of the physiological time series and the residual of the physiological reconstruction sequence under each classification probability. The second correlation is expressed as:

[0099]

[0100] Among them, G2 represents the second correlation, τ1 represents the standard deviation of the residuals of the physiological time series, τ0 represents the residual adjustment coefficient, which is used to constrain the values ​​of the residuals of the physiological time series and the residuals of the physiological reconstruction sequence in the first correlation, and determine whether the two residuals are correlated at this time, τ2 represents the standard deviation of the residuals of the physiological reconstruction sequence, and e represents the exponential constant.

[0101] When the classification corresponding to the maximum value of the first correlation and the second correlation is used as the output classification result, by determining the values ​​of the first correlation and the second correlation in sequence, first determine whether the physiological reconstruction sequence corresponding to the maximum values ​​of the first correlation and the second correlation is the same. If they are the same, output the physiological reconstruction sequence. The physiological reconstruction sequence obtained at this time can display the state of the current target object and the corresponding characteristics, and can show whether there is an abnormality in the current data of the target object after classification processing; if the physiological reconstruction sequences corresponding to the maximum values ​​of the first correlation and the second correlation are not the same, obtain the physiological reconstruction sequence corresponding to the maximum values ​​of the first correlation and the second correlation in the previous round of iteration, and output the physiological reconstruction sequence with the largest weighted average value of the first correlation and the second correlation at this time. The output physiological reconstruction sequence is the classification result required at this time.

[0102] At this time, when obtaining the previous round of iterations and this round of iterations, four physiological reconstruction sequences will appear, two of which are physiological reconstruction sequences with different maximum values ​​of the current first correlation and the second correlation, as well as two physiological reconstruction sequences in the previous round. At this time, the weighted average of these four sequences is calculated in turn, and the weights set are 0.4 for the first correlation and 0.6 for the second correlation. Then, the weighted average of the first correlation and the second correlation is calculated, and the part with the largest value at this time is used as the physiological reconstruction sequence to be output. At this time, the classification result of the corresponding sequence can be obtained, thereby improving the processing effect of data classification.

[0103] By continuously monitoring the patient's physiological data and classifying it in real time, once the model identifies risky data, it will issue an early warning based on the corresponding values ​​in the identified classification to remind medical professionals to take timely response measures.

[0104] It should be noted that the target audience includes the elderly as well as young people, or people with special needs.

[0105] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A human health monitoring method based on a wearable device, characterized in that: include: S1, obtaining physiological data of the target object; S2, adding health status labels to physiological data; The physiological data are converted into a time series according to the time of collection, which is recorded as a physiological time series; S3, performing data enhancement processing on the physiological time series, using the Bayesian method and information entropy to evaluate the physiological time series, and reconstructing the evaluated physiological time series into a physiological reconstruction sequence; S4, evaluating the physiological reconstruction sequence to determine the reconstruction effect of the physiological reconstruction sequence; S5, extracting features from the physiological reconstruction sequence, inputting the extracted features into a bidirectional long short-term memory network, and obtaining a classification result of the physiological reconstruction sequence.

2. A human health monitoring method based on a wearable device as claimed in claim 1, characterized in that: The data enhancement processing in step S3 includes: S31, perform wavelet transformation on the physiological time series, and use the Bayesian function to verify the posterior probability and prior probability of the physiological time series; S32, obtaining the information entropy of the physiological time series according to the calculated prior probability; S33, calculating compressed sensing parameters according to the acquired information entropy, and reconstructing the physiological time series using the compressed sensing parameters to obtain a physiological reconstructed sequence.

3. A human health monitoring method based on a wearable device as claimed in claim 1, characterized in that: The collected physiological data are added with corresponding health status labels; health status labels include normal, sub-health, hypertension, arrhythmia, and physical fatigue.

4. A human health monitoring method based on a wearable device as claimed in claim 3, characterized in that: The sparsely represented physiological time series can be expressed as: S=arg min∥W(ECG(t))∥0; Among them, ∥·∥0 represents the L0 norm, which is used to measure the number of non-zero elements in the vector. W() represents the wavelet transform, which transforms the physiological time series into the required sparse form; S is the sparsely represented physiological time series; W(ECG(t)) represents the physiological time series after wavelet transform.

5. A human health monitoring method based on a wearable device as claimed in claim 4, characterized in that: The adjustment strategy of the compressed sensing parameters of the compressed sensing algorithm can be expressed as: Among them, θ new represents the adjusted compressed sensing parameters, θ old represents the compressed sensing parameters before adjustment, α is the learning rate, which controls the speed of parameter update. By evaluating the uncertainty of S, the compressed sensing parameters are dynamically adjusted; Represents the gradient operator of the compressed sensing function.

6. A human health monitoring method based on a wearable device as claimed in claim 5, characterized in that: The physiological time series is reconstructed as follows: r k =y k -S k i k ; Among them, r k represents the residual of the physiological time series after the kth iteration, θ new,i represents the compressed sensing parameters after the kth iteration; y k represents the measured value after the kth iteration in the physiological time series; S k Represents the value of the element in the sparsely represented physiological time series at the kth iteration.

7. A human health monitoring method based on a wearable device as claimed in claim 6, characterized in that: The classification results are obtained by inputting the physiological reconstruction sequence, and selectively screening the temporal features of the physiological feature sequence through bidirectional long short-term memory network for bidirectional propagation; the extracted temporal features are input into the attention enhancement layer, and then classified using the Softmax classification function to obtain the classification probability of the physiological reconstruction sequence, and the corresponding classification result is determined according to the obtained classification probability.

8. A human health monitoring method based on a wearable device as claimed in claim 7, characterized in that: The obtained classification probability is expressed as: Among them, CA represents the final classification probability; Softmax represents the Softmax classification function, which is used to convert the output of the fully connected layer into a probability distribution; z b Represents the b-th element in the fully connected layer output vector z, b = 1, 2,, C; C represents the number of categories.

9. A human health monitoring method based on a wearable device as claimed in claim 8, characterized in that: After obtaining the classification probability, the classification result is obtained by successively judging the change value of the numerical value calculated by the classification probability, and determining the correlation between the classification probability of the physiological reconstruction sequence and the currently input physiological time series within a preset time period as the first correlation; based on the obtained first correlation, extracting the second correlation from the first correlation; and obtaining the output classification result based on the obtained first correlation and the second correlation.

10. A human health monitoring method based on a wearable device as claimed in claim 9, characterized in that: The implementation method for calculating the correlation between the classification probability of the physiological reconstruction sequence and the currently input physiological time series is to sort the physiological time series according to the classification probability, calculate the Pearson correlation coefficient between the physiological reconstruction sequence and the physiological time series under each classification probability in turn as the first correlation, calculate the second correlation with the residual of the physiological time series and the physiological reconstruction sequence, and take the classification corresponding to the maximum value between the first correlation and the second correlation as the output classification result.

Images