A wearable device, method, and system

Abstract

The application discloses a wearable device, method and system, and relates to the technical field of health monitoring.The wearable device comprises a physiological sensor and a processor.The physiological sensor is used for collecting a user physiological perception data stream, and outputting a physiological signal for generating enhanced measurement data in a short-time high-density resampling mode after a predictive measurement wake-up trigger.The processor is used for extracting features of the user physiological perception data stream to generate a physiological state measurement candidate probability distribution, performing uncertainty trend discrimination based on the candidate probability distribution to generate a predictive measurement wake-up trigger instruction, and completing low-power rough screening, physiological index refinement extraction and state convergence determination after waking up to output a target physiological state measurement conclusion, control a health prompt and generate a measurement record.The application improves the timeliness, accuracy and result reliability of abnormal physiological state measurement by identifying measurement value.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, and in particular to a wearable device, method, and system. Background Technology

[0002] With the continuous development of wearable electronics, micro-sensing, and embedded intelligent processing technologies, wearable devices have evolved from simple step counting and activity recording to comprehensive sensing terminals integrating physiological signal acquisition, health monitoring, and decision support. In existing technologies, physiological sensors such as photoplethysmography (PPG), electromyography (EMG), and body temperature sensors are widely used in scenarios such as heart rate monitoring, exercise load assessment, and body surface condition detection. Related research is gradually moving towards the joint analysis of multimodal physiological signals. Simultaneously, with the application of artificial intelligence at the edge, some wearable devices have begun to incorporate machine learning algorithms for automatic analysis of physiological signals to achieve preliminary identification of abnormal physiological states or health risk alerts.

[0003] However, existing wearable devices still have several key shortcomings. On the one hand, most existing physiological measurement schemes adopt constant sampling or passive triggering methods throughout the process, lacking forward-looking judgment on the trend of physiological state changes. This leads to entering the high sampling or high computing power processing stage before physiological changes are significant, resulting in unnecessary energy consumption and making it unsuitable for long-term continuous wear. On the other hand, most existing technologies lack a systematic characterization of the triggering sequence, synergistic relationship and uncertainty evolution process between different physiological modes in the process of multimodal physiological signal processing, which can easily lead to misjudgment or insufficient measurement reliability during the physiological state transition stage. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a wearable device to solve the problems of high power consumption and misjudgment during the transition phase.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a wearable device, which includes a physiological sensor and a processor;

[0008] The physiological sensor is used to collect the user's physiological perception data stream and outputs physiological signals to generate enhanced measurement data in a short-time high-density resampling manner after the predictive measurement wake-up trigger. The processor is used to extract features from the user's physiological perception data stream to generate a candidate probability distribution for physiological state measurement, perform uncertainty trend discrimination based on the candidate probability distribution to generate a predictive measurement wake-up trigger command, and after wake-up, complete low-computing-power coarse screening, refined extraction of physiological indicators and state convergence determination to output the target physiological state measurement conclusion, control health prompts and generate measurement records.

[0009] As a preferred embodiment of the wearable device described in this invention, the physiological sensor includes a photoplethysmography (PPG) sensor, an electromyography (EMG) sensor, and a body temperature sensor.

[0010] The user's physiological perception data stream includes pulse wave timing data collected by a photoplethysmography (PPG) sensor, electromyography (EMG) timing data collected by an EMG sensor, and body surface temperature timing data collected by a body temperature sensor.

[0011] In a preferred embodiment of the wearable device described in this invention, the processor is further configured to extract features from the user's physiological perception data stream and obtain a candidate probability distribution for physiological state measurements. The specific steps are as follows.

[0012] Based on user physiological perception data stream, perturbation contrastive training is performed through a self-supervised contrastive learning algorithm to generate multimodal physiological representation vectors;

[0013] Perform cross-modal trigger order verification on multimodal physiological representation vectors to generate trigger order deviation features;

[0014] Based on the characteristics of trigger sequence deviation, the user's physiological perception data stream is divided into a set of physiological segments by time window reconstruction;

[0015] The set of physiological segments is encoded with an uncertainty risk field and probability mapping is performed to obtain the candidate probability distribution of physiological state measurement.

[0016] In a preferred embodiment of the wearable device described in this invention, the processor is further configured to use an uncertainty trend discrimination method to determine whether the candidate probability distribution for physiological state measurement has shifted from a stable state to an abnormal physiological state measurement stage, and generate a predictive measurement wake-up trigger command. The specific steps are as follows.

[0017] Based on the candidate probability distribution of physiological state measurement, the dispersion of the probability distribution is calculated and sequential correlation is performed to construct an uncertain time series trajectory;

[0018] Based on uncertain time series trajectories, the trend direction and magnitude of change are analyzed using an uncertain trend discrimination method to obtain migration judgment results;

[0019] The migration determination result is converted into wake-up status parameters and encapsulated as a control instruction field to generate a predictive measurement wake-up trigger instruction.

[0020] In a preferred embodiment of the wearable device described in this invention, the processor is further configured to execute predictive measurement wake-up trigger instructions and perform low-computational-power coarse screening of the user's physiological perception data stream. The specific steps are as follows:

[0021] The trigger command is executed to switch the wearable device from a persistent sensing state to a low-power measurement preparation state;

[0022] Simplify and statistically analyze the user's physiological perception data stream, and generate a low-computing-power physiological change representation through trend compression;

[0023] Based on low-computation-power physiological change characterization, a fast consistency discrimination algorithm is used to determine whether the current physiological changes have further measurement value.

[0024] In a preferred embodiment of the wearable device described in this invention, the processor is further configured to generate enhanced measurement data, the specific steps of which are as follows:

[0025] When the current physiological changes are not of further measurement value, the physiological sensor should be kept in a resident low-sampling state.

[0026] When current physiological changes have further measurement value, perform short-term high-density resampling to generate enhanced measurement data.

[0027] In a preferred embodiment of the wearable device of the present invention, the processor is further configured to perform physiological index refinement extraction on the enhanced measurement data, and to perform state convergence determination to obtain the target physiological state measurement conclusion.

[0028] Based on enhanced measurement data, morphological feature extraction is performed on pulse wave time series data to extract cardiac rhythm-related indicators;

[0029] Activation fragmentation processing was performed on electromyographic time-series data to obtain muscle load-related indicators;

[0030] The time-series data of body surface temperature are subjected to trend extraction to obtain thermal change-related indicators, and a multimodal physiological indicator set is generated.

[0031] Cross-modal consistency constraints are applied to the set of multimodal physiological indicators to obtain a unified state evidence sequence. The dispersion trend of the state evidence is judged by a convergence decision algorithm to obtain the target physiological state measurement conclusion.

[0032] In a preferred embodiment of the wearable device described in this invention, the processor is further configured to control the wearable device to output health prompts and generate measurement records based on the measurement conclusions of the target physiological state. The specific steps are as follows.

[0033] The health prompt level value is calculated based on the measurement results of the target physiological state, and the health prompt method is obtained through the prompt intensity adaptive selection method.

[0034] Based on the health prompt method, the wearable device is controlled to output health prompts, and the target physiological state measurement results, health prompt level values ​​and health prompts are associated and encapsulated to generate measurement records.

[0035] Secondly, the present invention provides a wearable device system, comprising: a probability distribution generation module, which collects user physiological perception data streams and extracts features to obtain a candidate probability distribution for physiological state measurement; a wake-up determination module, which, based on the candidate probability distribution for physiological state measurement, uses an uncertainty trend discrimination method to determine whether a transition from a stable state to an abnormal physiological state measurement stage has occurred, and generates a predictive measurement wake-up trigger command; an enhancement control module, which executes the predictive measurement wake-up trigger command and performs low-computing-power coarse screening, and when it is confirmed that further measurement is needed, generates enhanced measurement data by performing short-term high-density resampling of the user physiological perception data stream and simultaneously collecting physiological signals; and a health prompt module, which performs refined extraction of physiological indicators and state convergence determination on the enhanced measurement data, obtains the target physiological state measurement conclusion, controls the wearable device to output health prompts, and generates a measurement record.

[0036] The beneficial effects of this invention are as follows: by identifying measurement value, it avoids the energy waste caused by long-term constant high sampling, and improves the temporal resolution of key physiological signals and the reliability of multimodal consistency judgment. While ensuring the long-term continuous wearing capability of wearable devices, it improves the timeliness, accuracy and reliability of abnormal physiological state measurement. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Fig. 1 This is a flowchart for wearable devices.

[0039] Fig. 2 A flowchart for generating predictive measurement wake-up trigger commands.

[0040] Fig. 3A flowchart for generating enhanced measurement data.

[0041] Fig. 4 A flowchart for generating measurement records. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] 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 phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0045] Reference Figs. 1-4 As one embodiment of the present invention, this embodiment provides a wearable device, including the following steps:

[0046] The probability distribution generation module collects user physiological perception data streams and extracts features to obtain candidate probability distributions for physiological state measurements.

[0047] The user's physiological perception data stream includes pulse wave timing data acquired by a photoplethysmography (PPG) sensor, electromyography (EMG) timing data acquired by an EMG sensor, and body surface temperature timing data acquired by a body temperature sensor.

[0048] Specifically, the photoplethysmography (PPG) sensor emits light into the subcutaneous blood vessel area using a light-emitting device, and the changes in reflected light intensity generated by the pulsation of blood volume are received by a photodetector. The changes in reflected light intensity are then converted into pulse wave time-series data that is continuously output over time.

[0049] Electromyography (EMG) sensors acquire changes in skin surface potential caused by muscle contraction and relaxation through electrodes attached to the skin surface. After amplification and filtering at the front end, the EMG time-series data is generated and continuously output over time.

[0050] The body temperature sensor comes into contact with the skin to acquire changes in skin surface temperature and converts these changes into time-series data of body surface temperature that is continuously output over time.

[0051] After recording the pulse wave time series data, electromyography time series data, and body surface temperature time series data according to a unified time reference, the user's physiological perception data stream is obtained.

[0052] Based on user physiological perception data streams, perturbation-based contrastive training is performed using a self-supervised contrastive learning algorithm to generate multimodal physiological representation vectors.

[0053] Specifically, based on the user's physiological perception data stream, according to the time sequence information recorded in the user's physiological perception data stream, the pulse wave time series data, electromyography time series data, and body surface temperature time series data are extracted into multiple continuous time windows within the same time range; within each continuous time window, pulse wave time series data segments, electromyography time series data segments, and body surface temperature time series data segments are read as reference time windows.

[0054] For pulse wave time series data, electromyography time series data, or body surface temperature time series data within the reference time window, the value of each sampling point is read. The product of the value of each sampling point and the scaling factor is used as the time series data after the amplitude is proportionally amplified or reduced. The time series data after amplitude scaling is used as the perturbation time window.

[0055] The baseline time window and the perturbation time window are input into the self-supervised contrastive learning algorithm. The pulse wave time series data, electromyography time series data and body surface temperature time series data in the baseline time window and the perturbation time window are read respectively. The data are then truncated to the same length and aligned with the time order to obtain a normalized time series set.

[0056] The normalized time series set is arranged into a multi-channel time series input according to the channel order of pulse wave time series data, electromyography time series data, and body surface temperature time series data. The mean, variation amplitude and adjacent difference of the multi-channel time series input are statistically analyzed by a self-supervised contrastive learning algorithm to generate a multimodal physiological representation vector.

[0057] It should be noted that the ratio of the maximum to the minimum sampling point value of pulse wave time series data, electromyography time series data, or body surface temperature time series data within the same time window is used as the scaling factor; the scaling factor is a proportional factor that amplifies or reduces the value of each sampling point within the time window, and the exemplary value range is as follows: Without altering the temporal sequence and morphological structure, the amplitude changes of the same physiological state at different intensity levels are simulated.

[0058] Self-supervised contrastive learning (SCL) is a feature learning method that does not rely on manually labeled data. It guides the feature encoding process to focus on the intrinsic structure of the data itself by constructing semantically consistent but differently represented data pairs from the same data source. SCL can learn stable representations that are sensitive to temporal patterns and changes but insensitive to amplitude perturbations, without introducing external labels, and can be used for subsequent physiological state analysis.

[0059] Perform cross-modal trigger order verification on the multimodal physiological representation vectors to generate trigger order deviation features.

[0060] Specifically, based on multimodal physiological representation vectors, the starting time position of the change from weak to strong is located from pulse wave time series data, electromyography time series data and body surface temperature time series data within the same time range; the actual triggering order is obtained by arranging the starting time positions of pulse wave time series data, electromyography time series data and body surface temperature time series data in sequence.

[0061] The actual triggering order is compared with the triggering order in the user's physiological perception data stream to generate triggering order deviation features.

[0062] Furthermore, the trigger sequence deviation feature refers to the order information of the pulse wave timing data start time position, the electromyography timing data start time position and the body surface temperature timing data start time position, as well as the deviation markers of whether the order information is consistent or inconsistent with the order information of the highest frequency occurrence in the time window of the user's physiological perception data stream.

[0063] Based on the characteristics of trigger sequence deviation, the user's physiological perception data stream is divided into a set of physiological segments by reconstructing through time windows.

[0064] Specifically, based on the trigger order deviation feature, the user's physiological perception data stream is divided into a continuous time window sequence according to the time order, and a corresponding trigger order deviation feature is associated with each continuous time window; the consistency in the trigger order deviation feature means that the order of the pulse wave time sequence data start time position, the order of the electromyography time sequence data start time position and the order of the body surface temperature time sequence data start time position within the continuous time window are the same as the order of the user's physiological perception data stream.

[0065] Inconsistency in the trigger sequence deviation feature refers to the difference between the order of the start time positions of the pulse wave time series data, the electromyography time series data, and the body surface temperature time series data within the continuous time window and their order in the user's physiological perception data stream.

[0066] In a continuous time window sequence, the time window in which the triggering order deviation feature changes from consistency to inconsistency is used as the segment start boundary, and the time window in which the triggering order deviation feature changes from inconsistency to consistency is used as the segment end boundary. The continuous time windows between the segment start boundary and the segment end boundary are merged to form a physiological segment. All physiological segments in the continuous time window sequence are counted to obtain the physiological segment set.

[0067] The set of physiological segments is encoded with an uncertainty risk field and probability mapping is performed to obtain the candidate probability distribution of physiological state measurement.

[0068] Specifically, based on the set of physiological segments, the processor reads the pulse wave time-series data, electromyography time-series data, and body surface temperature time-series data contained in each physiological segment in chronological order, and simultaneously reads the trigger sequence deviation characteristics corresponding to each physiological segment; within each physiological segment, the processor uses the starting time position of each pulse wave time-series data, electromyography time-series data, and body surface temperature time-series data as the starting point of change.

[0069] Based on the starting point of the change, extend backward to obtain the duration of the change, write the order of the triggering sequence deviation information into the same time axis, and retain the starting point of the change, the duration of the change, and the order of the deviation markers to obtain the uncertainty risk field code.

[0070] It should be noted that the uncertainty risk field coding maps the starting point of change, the duration of change, and the deviation of the order of change in physiological segments to the same time axis for structured expression. This allows the uncertainty distribution characteristics generated during the evolution of physiological state from stable to abnormal to be centrally characterized and used for subsequent probability statistics and trend judgment.

[0071] Based on the starting point of change, the duration of change, and the deviation markers of the order of change in the uncertainty risk field code, the distribution of the corresponding physiological state measurement candidate categories and the total number of candidate categories in the uncertainty risk field code are statistically analyzed to obtain the probability distribution of physiological state measurement candidates.

[0072] The wake-up determination module, based on the probability distribution of physiological state measurement candidates, uses an uncertainty trend discrimination method to determine whether a transition from a stable state to an abnormal physiological state measurement stage has occurred, and generates a predictive measurement wake-up trigger command.

[0073] Based on the candidate probability distribution of physiological state measurement, the dispersion of the probability distribution is calculated and sequential correlation is performed to construct an uncertain time series trajectory.

[0074] Specifically, based on the candidate probability distribution of physiological state measurements, the processor reads the candidate probability distribution and the total candidate probability value of physiological state measurements corresponding to consecutive time windows in chronological order. The ratio of the total candidate probability value of physiological state measurements to the total number of candidate categories of physiological state measurements is used as the baseline probability value of the uniform distribution. The dispersion value of the probability distribution for the corresponding time window is then calculated, expressed as:

[0075] ;

[0076] in, This represents the dispersion value of the probability distribution. For time windows Number of candidate categories for measuring physiological state within the body. In order to be with the first The number of physiological segments whose starting point of change, duration of change, and chronological order deviate from the label for each candidate category of physiological state measurement. For time window index value, For time windows The total number of physiological fragments included in the statistics For time windows Number of candidate categories for measuring physiological state within the body. The baseline probability value is a uniform distribution. Indicates time window Inner The probability proportion of each physiological state measurement candidate category in the uncertainty risk field coding statistics. The sequence number of the candidate category for physiological state measurement.

[0077] It should be noted that the formula for calculating the dispersion value of the probability distribution consists of the counting ratio within the same time window, and all terms are dimensionless proportional values. Therefore, the dimensions of the entire formula are consistent.

[0078] Arrange the probability distribution discrete values ​​sequentially in chronological order, and connect multiple probability distribution discrete values ​​of adjacent time windows in chronological order as a discrete sequence, which is used as an uncertain time series trajectory.

[0079] Based on uncertain time series trajectories, the trend direction and magnitude of change are analyzed using uncertain trend discrimination methods to obtain migration judgment results.

[0080] Specifically, based on the uncertain time-series trajectory, the processor reads the discrete sequence values ​​in the uncertain time-series trajectory in chronological order, and performs a comparison between each pair of adjacent time positions. The difference between the discrete sequence value of the later time position and the discrete sequence value of the previous time position is used as the comparison result. If the comparison result is positive, it indicates that the discreteness is increasing; if the comparison result is negative, it indicates that the discreteness is decreasing; and if the comparison result is zero, it indicates that the discreteness remains unchanged, thus obtaining the trend direction arranged in chronological order.

[0081] The absolute value of the difference in the discrete sequence values ​​of each pair of adjacent time positions is taken as the single-step change amplitude value; within the same continuous segment, the single-step change amplitude values ​​are summed up according to the maximum value to obtain the segment change amplitude.

[0082] Using the trend direction (reflected in the magnitude relationship of the uncertainty index between adjacent time positions) and the magnitude of change as input, the migration judgment result is "migration is successful" when the trend direction remains upward within a continuous segment and the magnitude of change within the segment meets the magnitude condition; the migration judgment result is "migration is not successful" when the trend direction sequence does not remain upward within a continuous segment or the magnitude of change within the segment does not meet the magnitude condition.

[0083] It should be noted that the amplitude condition refers to the requirement that the amplitude of change must reach the maximum value extracted from the recent set of amplitudes of uncertain time series trajectories, such as 10-60 time windows. This is used to limit the migration judgment result from being triggered by small fluctuations, so as to ensure that the migration judgment result is only valid when the amplitude of change exceeds the amplitude condition.

[0084] The migration determination result is converted into wake-up status parameters and encapsulated as a control instruction field to generate a predictive measurement wake-up trigger instruction.

[0085] Specifically, the migration determination result is read, and the target state of the wake-up state parameter is determined based on the migration determination result between migration success and migration failure. When the migration determination result is migration success, the wake-up state parameter is taken as the measurement wake-up state and simultaneously marked as measurement preparation. When the migration determination result is migration failure, the wake-up state parameter is taken as the resident sensing state and marked as maintaining low sampling.

[0086] After the wake-up status parameters are determined, the processor encapsulates the wake-up status parameters, measurement preparation flag or maintain low sampling flag in a fixed field order to generate a predictive measurement wake-up trigger instruction (including wake-up status parameter field and status flag field).

[0087] It should be noted that the measurement wake-up state refers to the working state that the wearable device enters in order to perform further measurements. The processor, based on the predictive measurement wake-up trigger instruction, switches the physiological sensor from resident low sampling to the measurement preparation mode required for short-term high-density resampling, and allows subsequent steps to perform low-computing coarse screening and enhanced measurement data generation.

[0088] The resident sensing state refers to the working state of wearable devices for continuous monitoring. The processor keeps the physiological sensors in a resident low sampling mode to collect the user's physiological sensing data stream and continuously generates the candidate probability distribution of physiological state measurement without triggering short-term high-density resampling.

[0089] The enhanced control module executes predictive measurement wake-up trigger commands and performs low-computing-power coarse screening. When it is confirmed that further measurement is needed, it generates enhanced measurement data by performing short-term high-density resampling of the user's physiological perception data stream and simultaneously acquiring physiological signals.

[0090] The trigger command is executed to switch the wearable device from a persistent sensing state to a low-power measurement preparation state.

[0091] Specifically, the processor reads the wake-up status parameter field and the status flag field in the order of the control instruction field in the predictive measurement wake-up trigger instruction; it uses the wake-up status parameter field as a low-power measurement preparation state switching request and the status flag field as a request to maintain resident low sampling acquisition.

[0092] The processor sends a sampling mode hold instruction to the physiological sensor, enabling the physiological sensor to continue collecting the user's physiological perception data stream in a resident sensing state and maintain the time sequence information record. At the same time, it writes a low-power measurement preparation status flag to the processor's internal scheduler to update the wearable device.

[0093] The user's physiological perception data stream is simplified and statistically analyzed, and a low-computing-power physiological change representation is generated through trend compression.

[0094] Specifically, the processor continuously reads the user's physiological perception data stream in a low-power measurement preparation state, and extracts adjacent consecutive time periods from the user's physiological perception data stream in chronological order as a statistical grid.

[0095] Within each statistical grid, the sampling point values ​​of pulse wave time series data, electromyography time series data, and body surface temperature time series data are traversed separately. The maximum and minimum values ​​of the pulse wave time series data are extracted, and the difference between the maximum and minimum values ​​of the pulse wave time series data is taken as the pulse wave fluctuation.

[0096] The maximum and minimum values ​​of the electromyography (EMG) time series data are extracted, and the difference between the maximum and minimum values ​​is taken as the EMG fluctuation. The values ​​of the first and last sampling points of the body surface temperature time series data are extracted, and the difference between the first and last sampling points of the body surface temperature time series data is taken as the body surface temperature change, thus obtaining simplified statistical results.

[0097] After the simplified statistical results are generated, the difference between the pulse wave fluctuation, electromyography fluctuation, and body surface temperature change of the next statistical grid and the previous statistical grid is used as the comparison result. A positive comparison result is marked as an upward marker, a negative comparison result is marked as a downward marker, and a zero comparison result is marked as a level marker. The upward marker, downward marker, and level marker are written into the compressed trend sequence (including the trend markers of pulse wave fluctuation, electromyography fluctuation, and body surface temperature change) in chronological order. The compressed trend sequence and the simplified statistical results together constitute a low-computation-power physiological change representation.

[0098] Based on low-computation-power physiological change characterization, a fast consistency discrimination algorithm is used to determine whether the current physiological changes have further measurement value.

[0099] Specifically, based on low-computing-power physiological change representation, the processor reads the trend markers of pulse wave fluctuation, electromyography fluctuation, and body surface temperature change in the compressed trend sequence in chronological order within the same time period; the fast consistency discrimination algorithm uses whether the trend markers of pulse wave fluctuation, electromyography fluctuation, and body surface temperature change are completely identical within the same time period as the consistency judgment condition.

[0100] When the trend markers for pulse wave fluctuation, electromyography fluctuation, and body surface temperature change are all either rising or falling within the same time period, a conclusion confirming the value of further measurement is output; when the trend markers for pulse wave fluctuation, electromyography fluctuation, and body surface temperature change exhibit different trend markers within the same time period, a conclusion denying the value of further measurement is output.

[0101] When the current physiological changes are not of further measurement value, the physiological sensor should be kept in a resident low-sampling state.

[0102] Specifically, the processor reads the further measurement value rejection conclusion and writes it into the wearable device's working state record. Based on the wearable device's working state record, the processor maintains a low-power measurement preparation state without switching to a measurement wake-up state, so as to keep the physiological sensor in a resident low-sampling state. In the resident low-sampling state, the physiological sensor continues to collect the user's physiological perception data stream and continues to record time sequence information.

[0103] When current physiological changes have further measurement value, perform short-term high-density resampling to generate enhanced measurement data.

[0104] Specifically, the conclusion of further measurement value confirmation will be written into the wearable device's working status record, and the wearable device will be switched from the low-power measurement preparation state to the measurement wake-up state.

[0105] Within the same time frame, the sampling frequency of the photoplethysmography (PPG) sensor is increased and high-density pulse wave time-series data is output. The sampling frequency of the electromyography (EMG) sensor is increased and high-density EMG time-series data is output. The sampling frequency of the body temperature sensor is increased and high-density body surface temperature time-series data is output. At the same time, the unified time reference of the user's physiological perception data stream is maintained, and the time sequence information is continued to be recorded and written into the high-density sampling point sequence.

[0106] The processor receives high-density pulse wave timing data, high-density electromyography timing data, and high-density body surface temperature timing data, and aligns and encapsulates them according to the time sequence information to form enhanced measurement data composed of high-density pulse wave timing data, high-density electromyography timing data, and high-density body surface temperature timing data.

[0107] The health alert module performs detailed extraction of physiological indicators from the enhanced measurement data, and obtains the target physiological state measurement conclusions through state convergence determination, thereby controlling the wearable device to output health alerts and generate measurement records.

[0108] Based on enhanced measurement data, morphological feature extraction is performed on pulse wave time series data to extract cardiac rhythm-related indicators.

[0109] Specifically, pulse wave time series data is read along the unified time sequence information recorded by the enhanced measurement data. The changes in values ​​of adjacent sampling points are traversed in the pulse wave time series data in chronological order to locate the starting position where the pulse wave time series data changes from stable to rapid rise. For example, after several consecutive sampling points have small fluctuations around 0.50, a continuous rising segment of 0.50--0.52--0.56--0.63 appears. The sampling point that starts from 0.50 and then becomes a continuous rapid rise is recorded as the starting position of the rise, which is used as the starting point of a single heartbeat.

[0110] Search for the peak position where the pulse wave time series data changes from rising to falling after the starting point of a single heartbeat. For example, if the pulse wave time series data after the starting point of a single heartbeat is 0.50--0.58--0.66--0.72--0.71--0.69, then the position corresponding to the inflection point of 0.72 where the value changes from rising to falling is recorded as the peak position, which is taken as the peak point of a single heartbeat.

[0111] Starting from the peak of a single heartbeat, the sampling points of the falling segment are read backwards to locate the end position of the falling segment where it enters a stable state. For example, the pulse wave time series data after the peak is 0.72--0.69--0.61--0.55--0.52--0.51--0.51. When the value changes from gradually decreasing to remaining continuously or approximately constant, the corresponding starting position is recorded as the end position of the stable falling segment, and the end position of the falling segment is taken as the end point of a single heartbeat. The time interval between the starting points of two adjacent heartbeats is recorded as the heartbeat interval, and the difference between the value of the peak of a single heartbeat and the value of the starting point is taken as the pulse wave amplitude.

[0112] The duration of the rise and fall of a single heartbeat are recorded as rise duration and fall duration, respectively. The heartbeat interval, pulse wave amplitude, rise duration and fall duration are summarized in chronological order to generate heart rhythm-related indicators.

[0113] Activation fragmentation processing was performed on the electromyography time series data to obtain muscle load-related indicators.

[0114] Specifically, the electromyography (EMG) time series data is read along the unified time sequence information recorded by the enhanced measurement data. Adjacent consecutive time periods are extracted from the EMG time series data in chronological order as activation discrimination time periods. Within each activation discrimination time period, the processor traverses the sampling points of the EMG time series data and summarizes the sum of the absolute values ​​of the sampling points as the EMG energy value. At the same time, the difference between the maximum and minimum values ​​of the sampling points of the EMG time series data is recorded as the EMG fluctuation amplitude value.

[0115] The electromyographic energy values ​​of adjacent activation discrimination time periods are compared in chronological order. When the amplitude of electromyographic fluctuation increases, the corresponding activation discrimination time period is marked as the activation time period, and adjacent activation time periods are merged into an activation segment in chronological order.

[0116] Read the start and end times of the activation segment to obtain the duration of activation; extract the maximum value of electromyographic energy within the activation segment as the peak load intensity; and count the number of activation discrimination time periods within the activation segment as the activation density.

[0117] The duration of activation, peak load intensity, and activation density were summarized in chronological order to form muscle load-related indicators.

[0118] The time-series data of body surface temperature are subjected to trend extraction to obtain thermal change-related indicators, and a multimodal physiological indicator set is generated.

[0119] Specifically, body surface temperature time series data is read along the unified time sequence information recorded by enhanced measurement data. Adjacent consecutive time periods are extracted from the body surface temperature time series data in chronological order as trend extraction time periods. Within each trend extraction time period, the difference between the value of the first sampling point and the value of the last sampling point of the trend extraction time period is taken as the temperature change.

[0120] The sampling point values ​​of the trend extraction time period are traversed and the difference between the maximum and minimum values ​​are recorded as the temperature fluctuation amplitude value. The temperature changes of adjacent trend extraction time periods are compared in chronological order. For example, when the temperature change changes from near zero to a sustained positive value, the corresponding trend extraction time period is marked as a warming trend segment, and when the temperature change changes from near zero to a sustained negative value, the corresponding trend extraction time period is marked as a cooling trend segment. Adjacent trend segments of the same type are merged into thermal change segments in chronological order.

[0121] Extract the start and end times of the thermal change segment to obtain the duration of the thermal change; extract the maximum value of the temperature fluctuation amplitude within the thermal change segment as the thermal fluctuation peak value, and use the duration of the thermal change and the thermal fluctuation peak value as thermal change-related indicators.

[0122] Within the same time frame, indicators related to thermal changes, heart rhythm, and muscle load are aligned and summarized in a unified time sequence to generate a multimodal physiological indicator set.

[0123] Cross-modal consistency constraints are applied to the set of multimodal physiological indicators to obtain a unified state evidence sequence. The dispersion trend of the state evidence is judged by a convergence decision algorithm to obtain the target physiological state measurement conclusion.

[0124] Specifically, based on the unified time sequence information of the multimodal physiological indicator set, heart rhythm-related indicators, muscle load-related indicators, and thermal change-related indicators within the same time range are read segment by segment as state evidence entries; within each state evidence entry, the change direction markers corresponding to the heart rhythm-related indicators, muscle load-related indicators, and thermal change-related indicators are extracted respectively.

[0125] The consistency of the change direction marker is used as the condition for the cross-modal consistency constraint to be valid, and the inconsistency of the change direction marker is used as the condition for the cross-modal consistency constraint to be invalid. When the cross-modal consistency constraint is invalid, the state evidence entry is treated as a conflict entry and written into the conflict source marker. The processor connects the state evidence entries retained in chronological order to form a unified state evidence sequence.

[0126] Based on the unified state evidence sequence, the change direction markers between adjacent state evidence items are extracted in chronological order and compared before and after execution. When the change direction markers of adjacent state evidence items remain consistent, the processor outputs a convergence judgment result of "convergence established". When the change direction markers of adjacent state evidence items switch frequently, the processor outputs a convergence judgment result of "convergence not established". The convergence established and convergence not established are uniformly expressed by a convergence judgment quantity. For example, the convergence judgment quantity is 1 when convergence is established and 0 when convergence is not established.

[0127] When convergence is achieved and the cross-modal consistency constraint is met, the output target physiological state measurement conclusion is marked as the abnormal physiological state measurement stage is established; when convergence fails and the cross-modal consistency constraint is not met, the output target physiological state measurement conclusion is marked as the abnormal physiological state measurement stage is not established. The establishment and non-establishment of the abnormal physiological state measurement stage are uniformly expressed by the abnormal physiological state measurement stage establishment judgment quantity. For example, when the abnormal physiological state measurement stage is established, the abnormal physiological state measurement stage judgment quantity is 1, and when the abnormal physiological state measurement stage is not established, the abnormal physiological state measurement stage judgment quantity is 0.

[0128] It should be noted that the convergence determination algorithm is used to determine whether the multimodal physiological indicators gradually tend to be stable and consistent over time; the convergence determination algorithm describes the process of physiological state moving from fluctuation to stability through temporal continuity and directional consistency.

[0129] The health alert level value is calculated based on the measurement results of the target physiological state, and the health alert method is obtained through an adaptive selection method of alert intensity.

[0130] Specifically, based on the measurement results of the target physiological state, the proportion of cross-modal consistency constraint establishment markers and convergence judgment results of the unified state evidence sequence within the same time range are statistically analyzed. The health warning level value is calculated by combining the establishment or non-establishment markers of the abnormal physiological state measurement stage, the proportion of cross-modal consistency constraint establishment markers, and the convergence judgment results. The expression is as follows:

[0131] ;

[0132] in, For health alert level values, For the consistency ratio, For the intensity of uncertainty, The total number of time windows. The total number of state evidence entries. The number of state evidence entries required to satisfy the conditions for holding cross-modal consistency constraints. The convergence criterion is used to determine whether the convergence is valid. This is a criterion established for the measurement phase of abnormal physiological states. For the first The probability distribution dispersion value corresponding to each time window.

[0133] Based on the ranking of health alert levels, the health alert method in the same measurement record is adjusted as the health alert level increases or decreases. For example, within the same measurement record formation period, when the health alert level is 0.30, the health alert method is vibration alert; when the health alert level is 0.55, the health alert method switches to screen alert; when the health alert level is 0.80, the health alert method switches to sound alert while simultaneously retaining vibration alert.

[0134] Based on the health prompt method, the wearable device is controlled to output health prompts, and the target physiological state measurement results, health prompt level values ​​and health prompts are associated and encapsulated to generate measurement records.

[0135] Specifically, based on the health prompt method, the prompt intensity information and prompt type information corresponding to the health prompt method are extracted, and the prompt intensity information and prompt type information are written into the health prompt control field of the wearable device.

[0136] The processor controls the wearable device to output health prompts based on the health prompt control field. The output duration and frequency are controlled by the prompt intensity information defined by the health prompt control field, and the output carrier, such as vibration prompts, sound prompts, and screen prompts, is controlled by the prompt type information defined by the health prompt control field. The processor also marks the health prompts and records the start time and end time of the health prompt output.

[0137] Based on the target physiological state measurement conclusion, health prompt level value, health prompt control field, health prompt execution flag, health prompt output start time and health prompt output end time, and using the time sequence information under the same unified time base as the association key, the target physiological state measurement conclusion field, health prompt level value field, health prompt method field, health prompt execution flag field and health prompt time information field are bound to the same measurement record to generate a measurement record.

[0138] This embodiment also provides a wearable device system, including: a probability distribution generation module, which collects user physiological perception data streams and extracts features to obtain a candidate probability distribution for physiological state measurement; a wake-up determination module, which, based on the candidate probability distribution for physiological state measurement, uses an uncertainty trend discrimination method to determine whether a transition from a stable state to an abnormal physiological state measurement stage has occurred, and generates a predictive measurement wake-up trigger command; an enhancement control module, which executes the predictive measurement wake-up trigger command and performs low-computing-power coarse screening, and when it is confirmed that further measurement is needed, generates enhanced measurement data by performing short-term high-density resampling of the user physiological perception data stream and simultaneously collecting physiological signals; and a health prompt module, which performs refined extraction of physiological indicators and state convergence determination on the enhanced measurement data, obtains the target physiological state measurement conclusion, controls the wearable device to output health prompts, and generates measurement records.

[0139] This embodiment also provides a computer device suitable for wearable devices, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the wearable device as proposed in the above embodiment.

[0140] In summary, this invention, by identifying measurement value, avoids the energy waste caused by long-term constant high sampling, and improves the temporal resolution and reliability of multimodal consistency judgment of key physiological signals. While ensuring the long-term continuous wearability of wearable devices, it improves the timeliness, accuracy and reliability of abnormal physiological state measurement results.

[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A wearable device, characterized in that, The wearable device includes: Physiological sensors and processors; The physiological sensor is used to collect user physiological perception data streams and output physiological signals to generate enhanced measurement data in a short-time high-density resampling manner after the predictive measurement wake-up trigger. The processor is used to extract features from the user's physiological perception data stream to generate a candidate probability distribution for physiological state measurement, perform uncertainty trend discrimination based on the candidate probability distribution to generate a predictive measurement wake-up trigger command, and after wake-up, complete low-computing-power coarse screening, refined extraction of physiological indicators and state convergence determination to output the target physiological state measurement conclusion, control health prompts and generate measurement records.

2. The wearable device as described in claim 1, characterized in that: The physiological sensors include a photoplethysmography (PPG) sensor, an electromyography (EMG) sensor, and a body temperature sensor. The user's physiological perception data stream includes pulse wave timing data collected by a photoplethysmography (PPG) sensor, electromyography (EMG) timing data collected by an EMG sensor, and body surface temperature timing data collected by a body temperature sensor.

3. The wearable device as described in claim 2, characterized in that: The processor is also used to extract features from the user's physiological perception data stream and obtain the candidate probability distribution of physiological state measurements. The specific steps are as follows. Based on user physiological perception data stream, perturbation contrastive training is performed through a self-supervised contrastive learning algorithm to generate multimodal physiological representation vectors; Perform cross-modal trigger order verification on multimodal physiological representation vectors to generate trigger order deviation features; Based on the characteristics of trigger sequence deviation, the user's physiological perception data stream is divided into a set of physiological segments by time window reconstruction; The set of physiological segments is encoded with an uncertainty risk field and probability mapping is performed to obtain the candidate probability distribution of physiological state measurement.

4. The wearable device as described in claim 3, characterized in that: The processor is further configured to use an uncertainty trend discrimination method to determine whether the candidate probability distribution for physiological state measurement has shifted from a stable state to an abnormal physiological state measurement stage, and generate a predictive measurement wake-up trigger instruction. The specific steps are as follows. Based on the candidate probability distribution of physiological state measurement, the dispersion of the probability distribution is calculated and sequential correlation is performed to construct an uncertain time series trajectory; Based on uncertain time series trajectories, the trend direction and magnitude of change are analyzed using an uncertain trend discrimination method to obtain migration judgment results; The migration determination result is converted into wake-up status parameters and encapsulated as a control instruction field to generate a predictive measurement wake-up trigger instruction.

5. The wearable device as described in claim 4, characterized in that: The processor is also used to execute predictive measurement wake-up trigger instructions and perform low-computational-power coarse screening of the user's physiological perception data stream. The specific steps are as follows. The trigger command is executed to switch the wearable device from a persistent sensing state to a low-power measurement preparation state; Simplify and statistically analyze the user's physiological perception data stream, and generate a low-computing-power physiological change representation through trend compression; Based on low-computation-power physiological change characterization, a fast consistency discrimination algorithm is used to determine whether the current physiological changes have further measurement value.

6. The wearable device as described in claim 5, characterized in that: The processor is also used to generate enhanced measurement data, the specific steps of which are as follows. When the current physiological changes are not of further measurement value, the physiological sensor should be kept in a resident low-sampling state. When current physiological changes have further measurement value, perform short-term high-density resampling to generate enhanced measurement data.

7. The wearable device as described in claim 6, characterized in that: The processor is also used to perform physiological index refinement extraction on the enhanced measurement data, and to determine state convergence to obtain the target physiological state measurement conclusion. Based on enhanced measurement data, morphological feature extraction is performed on pulse wave time series data to extract cardiac rhythm-related indicators; Activation fragmentation processing was performed on electromyographic time-series data to obtain muscle load-related indicators; The time-series data of body surface temperature are subjected to trend extraction to obtain thermal change-related indicators, and a multimodal physiological indicator set is generated. Cross-modal consistency constraints are applied to the set of multimodal physiological indicators to obtain a unified state evidence sequence. The dispersion trend of the state evidence is judged by a convergence decision algorithm to obtain the target physiological state measurement conclusion.

8. The wearable device as described in claim 7, characterized in that: The processor is also used to control the wearable device to output health prompts and generate measurement records based on the measurement results of the target physiological state. The specific steps are as follows. The health prompt level value is calculated based on the measurement results of the target physiological state, and the health prompt method is obtained through the prompt intensity adaptive selection method. Based on the health prompt method, the wearable device is controlled to output health prompts, and the target physiological state measurement results, health prompt level values ​​and health prompts are associated and encapsulated to generate measurement records.

9. A method for controlling a wearable device, characterized in that, include: Automatic control is achieved through the wearable device as described in any one of claims 1 to 8.

10. A wearable device system, characterized in that, include: Smart terminals, cloud servers, and wearable devices as described in any one of claims 1 to 8.

Images