Human state monitoring method and system based on low-power data feature extraction

Abstract

The application discloses a human state monitoring method and system based on low-power data feature extraction, and relates to the technical field of state monitoring, and comprises the following steps: collecting motion signals and physiological signals of a target human body to form an initial low-power data sequence; performing feature migration processing on the initial low-power data sequence to generate a migration coupling feature sequence; based on the migration coupling feature sequence, transient features sensitive to micro states are enhanced through a dynamic weighting strategy to obtain an enhanced micro state feature sequence; a feature subset is dynamically selected from the enhanced micro state feature sequence to generate a power consumption optimized feature sequence; and the power consumption optimized feature sequence is input into a human state judgment model to obtain a human state judgment result. Through coupling feature completion and sensitive feature enhancement of the motion signals and the physiological signals, the application realizes real-time judgment of human micro states, so as to solve the problem of insufficient micro state recognition precision caused by loss of nonlinear coupling information under a low-power condition.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring technology, and more specifically, to a human condition monitoring method and system based on low-power data feature extraction. Background Technology

[0002] With the development of wearable devices and intelligent health monitoring technologies, human condition monitoring has been widely used in scenarios such as sports and health, chronic disease management, and elderly care. Existing technologies typically analyze human condition by collecting motion signals (such as acceleration and angular velocity) and physiological signals (such as heart rate, respiration, and skin conductance), and rely on high sampling rates to capture minute changes in condition, thereby enabling the assessment of exercise patterns, fatigue levels, and health risks.

[0003] However, traditional methods face significant limitations under low-power sampling conditions. On the one hand, low sampling rates make it difficult to accurately acquire nonlinear transient coupling information between motion signals and physiological signals; on the other hand, since micro-state changes typically exhibit short-term nonlinear characteristics, insufficient sampling can lead to the loss of crucial information, resulting in decreased accuracy in micro-state recognition and limited real-time performance. This problem is particularly prominent in long-term monitoring of wearable devices or in low-power environments, where traditional data processing and feature extraction methods struggle to balance low power consumption with high accuracy.

[0004] The above-disclosed technical solutions have at least the following technical problems: In the prior art, human state monitoring relies on high sampling rate motion signals and physiological signals to capture micro-state changes. However, under low power sampling conditions, due to the limited sampling rate, it is difficult to obtain the nonlinear transient coupling information between motion signals and physiological signals, resulting in the loss of micro-state information, thereby reducing the real-time recognition capability of micro-state changes.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a human state monitoring method and system based on low-power data feature extraction. By collecting human motion signals and physiological signals to form a low-power data sequence, and performing feature transfer, dynamic weighting to enhance micro-state sensitive features in the sampling gap region, dynamically selecting feature subsets based on power consumption level, and inputting the human state judgment model for real-time identification, the method solves the problem of low micro-state recognition accuracy caused by the loss of nonlinear coupling information under low-power conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] On the one hand, the human state monitoring method based on low-power data feature extraction includes the following steps: collecting motion signals and physiological signals of the target human body to form an initial low-power data sequence; performing feature transfer processing on the initial low-power data sequence to generate a transfer-coupled feature sequence; based on the transfer-coupled feature sequence, enhancing transient features sensitive to micro-states through a dynamic weighting strategy to obtain an enhanced micro-state feature sequence; dynamically selecting a feature subset from the enhanced micro-state feature sequence according to the current power consumption level and sampling conditions to generate a power-optimized feature sequence; and passing the power-optimized feature sequence through a human state determination model to obtain the human state determination result.

[0009] In a preferred embodiment, the acquisition of motion and physiological signals of the target human body to form an initial low-power data sequence specifically involves: determining the energy consumption threshold range of the target node based on the energy consumption and information mapping relationship of the multi-mode sensing node; asynchronously sampling the motion and physiological signals based on the energy consumption threshold range to generate a cross-modal time mapping matrix, calculating the transient response similarity of the sampled segments to select preliminary sampling blocks; performing energy adaptive compression coding on the preliminary sampling blocks, setting the sampling points with information entropy contribution below a preset threshold to zero, and generating compressed low-power joint data segments; and splicing multiple low-power joint data segments in chronological order and performing interpolation smoothing to form the initial low-power data sequence.

[0010] In a preferred embodiment, the asynchronous sampling of motion signals and physiological signals based on an energy consumption threshold range specifically involves: continuously acquiring motion signals in a first sampling mode, wherein the motion signals include at least one of acceleration and angular velocity components; acquiring physiological signals in a second sampling mode, wherein the second sampling mode is triggered by a rate of change event of the motion signals; and establishing a cross-modal time mapping matrix for calibrating the temporal relationship between motion signals and physiological signals.

[0011] In a preferred embodiment, the step of performing feature transfer processing on the initial low-power data sequence to generate a transfer-coupled feature sequence specifically involves: identifying the sampling gap region of each modal signal in the initial low-power data sequence and determining its temporal boundary; constructing a cross-modal mutual mapping model with motion signal as the driving variable and physiological signal as the target variable, and calculating the transfer weight; and performing transfer inference and calibration on the target modal signal based on the transfer weight to form a transfer-coupled feature sequence.

[0012] In a preferred embodiment, identifying the sampling gap region of each modal signal in the initial low-power data sequence includes: jointly identifying the sampling gap region caused by low-power sampling constraints based on the signal response disappearance and the device's low-power state; and dynamically inferring the temporal boundary of the sampling gap region by referring to the multimodal co-sampling points before and after the sampling gap region and combining a preset time offset coefficient and historical propagation weight.

[0013] In a preferred embodiment, the step of enhancing the transient features sensitive to microstates based on the migration coupling feature sequence through a dynamic weighting strategy to obtain an enhanced microstate feature sequence specifically involves: normalizing the migration coupling feature sequence to form a cross-modal transient response vector, and obtaining the weight vectors for each mode based on historical sample analysis; dynamically correcting the weight vectors according to real-time energy consumption constraints and signal fluctuations to obtain a real-time weighted matrix; and performing weighted fusion processing on the migration coupling feature sequence and the real-time weighted matrix to output the enhanced microstate feature sequence.

[0014] In a preferred embodiment, the modal weight vector is obtained through the following steps: calculating the temporal gradient rate of change of each modal signal within a sliding time window; determining the modal sensitivity factor based on the mutual information value between the gradient rate of change and the microstate label; and generating the modal weight vector by normalizing the sensitivity factor.

[0015] In a preferred embodiment, the step of dynamically selecting a subset of features from the enhanced microstate feature sequence to generate a power-optimized feature sequence based on the current power consumption level and sampling conditions specifically involves: constructing a mapping relationship between features and power consumption; obtaining the current power consumption state of the device to determine the allocable power consumption budget and screening a set of candidate features that meet the budget constraints; calculating the temporal correlation between each feature and the target microstate and performing multi-factor comprehensive ranking; and selecting a subset of features based on the ranking results and performing feature fusion to generate a power-optimized feature sequence.

[0016] In a preferred embodiment, obtaining the human state determination result by passing the power consumption optimization feature sequence through the human state determination model specifically involves: calculating the instantaneous response weight of each feature component of the power consumption optimization feature sequence through the human state determination model, and generating a dynamic representation vector based on the weight; calculating the micro-state confidence based on the dynamic representation vector, and performing a re-determination process on the low confidence results to output a stable human state determination result; the re-determination process includes verifying the current state using historical determination results or supplementary features.

[0017] On the other hand, the human state monitoring system based on low-power data feature extraction includes the following modules: a low-power data acquisition module, used to acquire motion signals and physiological signals of the target human body to form an initial low-power data sequence; a nonlinear coupling migration prediction module, used to perform feature migration processing on the initial low-power data sequence to generate a migration coupling feature sequence; a micro-state enhancement feature generation module, used to enhance transient features sensitive to micro-states based on the migration coupling feature sequence through a dynamic weighting strategy to obtain an enhanced micro-state feature sequence; a power consumption feature screening module, used to dynamically select a feature subset from the enhanced micro-state feature sequence according to the current power consumption level and sampling conditions to generate a power consumption optimized feature sequence; and a human state determination module, used to pass the power consumption optimized feature sequence through a human state determination model to obtain the human state determination result.

[0018] The technical effects and advantages of the human body state monitoring method and system based on low-power data feature extraction of this invention are as follows:

[0019] 1. This invention acquires human motion and physiological signals to form an initial low-power data sequence. Based on this low-power data sequence, feature transfer is performed on the sampling gap region to generate a migration coupling feature sequence. Furthermore, a dynamic weighting strategy is used to enhance micro-state sensitive transient features. A feature subset is dynamically selected based on device power consumption level and sampling conditions to generate a power-optimized feature sequence. Finally, this sequence is input into a human state determination model to achieve real-time micro-state recognition. This method can retain nonlinear coupling information and enhance key feature responses under low-power conditions, improving the accuracy of micro-state determination. Simultaneously, it achieves adaptive power consumption management, extends device battery life, and ensures monitoring stability. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the human body state monitoring method based on low-power data feature extraction according to the present invention.

[0021] Figure 2 This is a schematic diagram of the human condition monitoring system based on low-power data feature extraction according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, Figure 1 The present invention provides a human state monitoring method based on low-power data feature extraction, comprising the following steps:

[0024] S1, collects motion and physiological signals of the target human body to form an initial low-power data sequence;

[0025] In this embodiment, the process of collecting motion and physiological signals from the target human body to form an initial low-power data sequence specifically involves:

[0026] S11. Establish an energy consumption-information mapping table for multi-mode sensing nodes. Based on the energy consumption and information mapping relationship of multi-mode sensing nodes, monitor the instantaneous energy consumption changes of each node in real time and determine the energy consumption threshold range of the target node. The energy consumption threshold range is used to limit the dynamic adjustment range of subsequent sampling cycles to avoid excessive power consumption due to oversampling or signal distortion due to undersampling.

[0027] S12. Based on the energy consumption threshold range, asynchronous trigger sampling is performed on motion signals and physiological signals respectively:

[0028] The motion signal is periodically sampled to obtain the acceleration and angular velocity components.

[0029] Event-driven sampling of physiological signals is used, triggering the acquisition of heart rate, skin conductance, or respiratory signals only when the rate of change of motion signals exceeds a preset threshold.

[0030] A cross-modal time mapping matrix is ​​established during asynchronous sampling to calibrate the relative timing relationship between each signal sampling point;

[0031] S13. Based on the cross-modal time mapping matrix, calculate the transient response similarity of each sampling segment to obtain a preliminary sampling block that reflects the nonlinear coupling characteristics of motion-physiology; the transient response similarity is obtained by the correlation gain function within the sliding time window and is used to measure the transient mutual excitation between different modal signals.

[0032] S14. Perform energy adaptive compression coding on the initial sampling block, set the sampling points with information entropy contribution below the preset threshold to zero, and generate compressed low-power joint data segments.

[0033] S15. The consecutive low-power joint data segments are spliced ​​together in chronological order and smoothed by interpolation to form an initial low-power data sequence.

[0034] It should be noted that directly increasing the sampling frequency under low-power conditions will significantly increase the energy consumption burden of the sensing nodes, leading to a decrease in the battery life of the monitoring equipment; while simply reducing the sampling rate will cause the loss of nonlinear transient coupling information between human motion signals and physiological signals, making it difficult to accurately characterize micro-state changes. Therefore, this embodiment introduces energy consumption sensing, adaptive sampling, and cross-modal synchronization mechanisms in the acquisition phase to retain key coupling characteristics under power-constrained conditions. By modeling the real-time energy consumption status of the sensing nodes, the mapping relationship of information acquisition capabilities under different power consumption levels is obtained. Low-power monitoring equipment exhibits large dynamic fluctuations in energy consumption under different voltage and temperature conditions, making stable operation difficult with a fixed sampling period. Technical advantages: By limiting the adjustment range of the sampling period through an energy consumption threshold, a quantitative balance between energy consumption and information acquisition is achieved, avoiding power waste due to oversampling or signal distortion due to undersampling.

[0035] S2, perform feature transfer processing on the initial low-power data sequence to generate a transfer-coupled feature sequence;

[0036] In this embodiment, the step of performing feature transfer processing on the initial low-power data sequence to generate a transfer-coupled feature sequence specifically involves:

[0037] S21. Based on the initial low-power data sequence, identify the sampling gap region of each modal signal, and determine the timing boundary corresponding to the missing segment according to the cross-modal time mapping matrix;

[0038] The sampling gap region is used to represent the uncollected interval due to low-power sampling constraints, and the determination of the timing boundary provides a constraint range for subsequent migration calculations.

[0039] S22. Within the temporal boundary, construct a cross-modal mutual mapping model, using the transient changes of motion signals as the driving variable and the response trend of physiological signals as the target variable, and calculate the driving-response transfer weights.

[0040] The migration weights are obtained using the minimum energy consumption deviation criterion, ensuring that the energy consumption cost of migration calculation does not exceed a preset energy consumption threshold.

[0041] S23. Based on the transfer weights, perform directional transfer inference on the target modal signal in the sampling gap region:

[0042] When there are abrupt changes in motion signals but no physiological signals, the response trend of the corresponding physiological signals can be predicted by using the transient gradient information of the driving variables.

[0043] When physiological signals fluctuate but motion signals are missing, back-transfer inference can be used to determine the potential amplitude changes in motion modes.

[0044] The bidirectional migration results are corrected for consistency constraints to obtain the coupled complete fragment.

[0045] S24. Perform cross-modal similarity backtracking calibration on the coupled completion fragment, calculate the mutual information gain of the front and back windows, and dynamically correct the transfer results based on the gain threshold, thereby eliminating the drift error of transfer inference in the low sampling interval.

[0046] S25. The calibrated coupling completion fragments are embedded into the original low-power data sequence in chronological order to form a migration coupling feature sequence with continuous structure and consistent information coupling.

[0047] S21, based on the initial low-power data sequence, identifies the sampling gap region of each modal signal, and determines the temporal boundary corresponding to the missing segment according to the cross-modal time mapping matrix, specifically as follows:

[0048] Based on the cross-modal time mapping matrix, the sampling time points of motion signals and physiological signals are uniformly indexed to form a global time index table;

[0049] Within each time unit of the global time index table, if a modal signal has not been updated for multiple consecutive sampling periods, and the corresponding transient response similarity decrease rate exceeds a preset threshold (indicating that the mode has lost its coupling response), and the energy consumption state of this time period is in the low power suppression range (indicating that the signal itself is not stationary but that sampling is paused), then this time period is determined to be the sampling gap region of the target mode.

[0050] The advantage of this identification logic is that the system does not simply detect missing signals through time intervals, but rather determines the sampling gap by combining "disappearance of coupling characteristics + power consumption constraint state", thus accurately distinguishing between "signal that is truly still" and "signal that is missing" under low sampling rate conditions.

[0051] The sampling time points of adjacent modal signals in the sampling gap region are found by using a cross-modal time mapping matrix, and the time boundaries are dynamically inferred using their mapping weights. Specifically, this includes:

[0052] The most recent multimodal co-sampling point before the start of the sampling interval is used as the reference point;

[0053] The first responsive sampling point after the end of the sampling interval is taken as the termination point;

[0054] The inference boundary time of the missing segment is calculated by using the time offset coefficient and the relevant propagation weight in the mapping matrix; the time offset coefficient and propagation weight can be obtained by fitting the minimum variance of historical co-sampled data, and are used to reflect the response delay relationship between different modal signals.

[0055] The identified sampling gap regions and their corresponding timing boundaries are recorded in the low-power data sequence index in the form of labeled vectors.

[0056] The annotation vector includes a time index, a modality identifier, and a gap boundary triplet.

[0057] S3, based on the transfer coupling feature sequence, enhances the transient features sensitive to microstates through a dynamic weighting strategy to obtain the enhanced microstate feature sequence;

[0058] In this embodiment, the enhanced micro-state feature sequence is obtained by using a dynamic weighting strategy to enhance transient features sensitive to micro-states based on the migration coupling feature sequence. Specifically:

[0059] S31. Normalize the local feature segments from different signal modes (motor, physiological, etc.) in the migration coupling feature sequence at the same time scale to obtain a cross-modal transient response vector set; each response vector contains acceleration change amplitude, physiological signal transient phase shift and response delay factor, etc., to characterize the dynamic coupling relationship in the process of human microstate occurrence.

[0060] S32. Based on the historical sample set of migration coupling feature sequences, perform correlation sensitivity analysis on the transient response vectors of each mode to obtain a weight vector that reflects the intensity of the mode's response to micro-state changes.

[0061] The weight vector is obtained through the following steps:

[0062] Calculate the rate of change of the temporal gradient of each modal signal within the sliding time window;

[0063]

[0064] The modal sensitivity factor is determined based on the mutual information value between the gradient rate of change and the microstate label.

[0065]

[0066] The sensitivity factors are normalized to obtain the initial set of weighted coefficients;

[0067]

[0068] in, Let m be the rate of change of the gradient of mode m at time t. This is the normalized transient response vector. The width of the sliding time window (number of sampling points) is used to calculate the local gradient. Let be the mutual information value between mode m and its microstates. , These are the marginal probability distributions of the gradient and the label, respectively. The joint probability distribution of gradient and microstate label. The initial sensitivity weights for mode m, This represents the total number of modes in the migration coupling feature sequence.

[0069] S33. Based on energy consumption constraints and real-time signal fluctuations, the initial weighting coefficient set is dynamically corrected to obtain a real-time weighting matrix. If the device is in a high power consumption range, the weight of low-sensitivity modes is automatically reduced to suppress invalid signal amplification. If the device is in a low power consumption range and a state change is detected, the weight of modes with high transient correlation is temporarily increased to enhance the sensitivity of micro-state change detection.

[0070] S34. The migration coupling feature sequence is multiplied with the real-time weighted matrix by mode point by point and then smoothed in the time domain to form an enhanced micro-state feature sequence. This enhanced micro-state feature sequence improves the response strength to short-term features such as small attitude changes, physiological mutations or stress responses while maintaining low power consumption constraints.

[0071] S35. Calculate the discrimination gain of features before and after enhancement based on the historical micro-state sample library. When the discrimination gain exceeds the preset threshold, confirm that the dynamic weighting strategy is effective and update it to the subsequent online recognition model.

[0072] The real-time weighted matrix is ​​calculated using the following formula:

[0073]

[0074] The enhanced micro-state feature sequence is specifically as follows:

[0075]

[0076]

[0077] in, Let be the real-time weighting matrix of mode m at time point t. As the initial sensitivity weight, This is a weighted adjustment coefficient (adjusted according to actual needs). The transient response correlation function, Real-time power consumption of the device. The maximum allowable power consumption of the device. This is a time-smoothed enhanced micro-state feature sequence. The sliding window size is used for time smoothing. These are the weighted transient eigenvalues. This is the normalized transient response vector.

[0078] The specific formula for calculating the discrimination gain is as follows:

[0079]

[0080] in, To enhance the discriminative gain of features relative to the original features, To enhance the micro-state feature sequence, This is the original migration coupling feature sequence. The sample label indicating the existence of a microstate. It is a variance function used to quantify and enhance the discriminative ability of features.

[0081] S4. Based on the current power consumption level and sampling conditions, dynamically select a subset of features from the enhanced microstate feature sequence to generate a power consumption optimized feature sequence;

[0082] The step of dynamically selecting a feature subset from the enhanced micro-state feature sequence based on the current power consumption level and sampling conditions to generate a power-optimized feature sequence is as follows:

[0083] S41. Based on the signal sampling frequency and computational complexity corresponding to each feature in the enhanced microstate feature sequence, construct a feature power consumption mapping matrix to characterize the energy consumption of different features under the current sampling conditions.

[0084] S42. Obtain the current power consumption level of the device and the remaining power information of the sensing nodes, calculate the allocable power consumption budget, and filter the enhanced microstate feature sequence according to the power consumption mapping matrix to obtain a set of candidate features that meet the power consumption budget constraints.

[0085] S43. Based on the candidate feature set and combined with the cross-modal time mapping matrix, calculate the temporal correlation between each feature and the target micro-state, and sort them according to the joint priority of correlation and power consumption weight.

[0086] S44. Based on the ranking results, select a subset of features with low power consumption and sensitivity to micro-state changes, and perform temporal interpolation and normalization fusion among similar features to generate a power-optimized feature sequence.

[0087] S5 uses the power consumption optimization feature sequence through the human body state determination model to obtain the human body state determination result.

[0088] In this embodiment, the step of obtaining the human state determination result by passing the power consumption optimization feature sequence through the human state determination model is specifically as follows:

[0089] S51. Input the power consumption optimization feature sequence into the pre-trained human state determination model;

[0090] S52. In the feature interaction layer of the model, the input features are mapped between layers based on the nonlinear decision function, and the instantaneous response weights of each feature component to the target micro-state are calculated.

[0091] The nonlinear decision function is used to capture high-order coupling relationships between features and dynamically adjust the decision threshold based on historical feature distribution.

[0092] S53. Using the instantaneous response weight as a dynamic discriminant factor for micro-state recognition, the input sequence is weighted according to time-by-time features to generate a dynamic representation vector for state determination.

[0093] S54. Calculate the micro-state confidence level in the output layer based on the dynamic representation vector, and perform a re-judgment process on the state results with the output confidence level lower than the preset threshold to obtain a stable human state judgment result.

[0094] S55. Output the final judgment result to the upper-level monitoring module to realize real-time identification of human micro-state and adaptive closed-loop control of power consumption.

[0095] The training process of the human body state determination model specifically includes the following steps:

[0096] Based on the initial low-power data sequence and its feature sequence enhanced by transfer coupling, multiple sets of training samples are generated according to different power levels and sampling conditions. Each set of samples includes motion signal components, physiological signal components and corresponding manually labeled human state labels, which are used to characterize the sparse distribution of features under different power scenarios.

[0097] During model training, power consumption constraints are set to limit the energy consumption of the model in terms of feature dimensions and computational complexity, and a phased training strategy is executed to obtain the human state determination model.

[0098] The first stage involves pre-training a model structure based on the full feature dataset to obtain the basic mapping relationships between features.

[0099] In the second stage, transfer training is performed under low-power sampling conditions, activating only the network channels related to enhancing micro-state features, and suppressing redundant computation through channel sparsity constraints.

[0100] Furthermore, the human state monitoring method based on low-power data feature extraction also includes: dynamically adjusting the low-power sampling strategy, including the sampling interval and sensor combination, according to the state determination result, to enhance the ability to capture micro-state changes while maintaining low power consumption, and feeding the adjusted strategy back to step S1 for the next round of data acquisition, specifically:

[0101] Based on the output human state determination results, the state change rate parameter is extracted to measure the intensity of changes in human state between adjacent time points;

[0102] Based on the state change rate parameter, the sampling adjustment factor is dynamically calculated, and the sampling interval of the multimode sensing node is updated.

[0103] Simultaneously calculate the sensor combination weight vector The combination of sensors participating in sampling is dynamically adjusted. When a low-confidence mode fluctuates continuously over multiple cycles, the sampling frequency of that mode is automatically reduced or the node is turned off to further reduce overall power consumption.

[0104] The updated sampling interval and sensor combination weights are fed back to step S1 to rebuild the energy consumption-information mapping table and execute the next round of asynchronous sampling process, thereby realizing the adaptive closed-loop optimization of power consumption for human body status monitoring.

[0105] The sampling adjustment factor is specifically:

[0106]

[0107] The sampling interval for updating the multimode sensing node is specifically as follows:

[0108]

[0109]

[0110]

[0111] in, For sampling adjustment factor, The parameter is the rate of state change. This is a preset smoothing coefficient used to prevent over-adjustment caused by short-term fluctuations. The sampling interval for the updated multimode sensing nodes. This serves as the basic sampling interval for the sensor nodes. For each weighted component, The confidence contribution value for the i-th sensor mode. The total number of sensors, Let be the power consumption weighting factor for the i-th sensor mode. Let be the variance of the feature sequence corresponding to the i-th mode, used to reflect the stability of the mode's features. The smaller the variance, the higher the reliability of the mode. This is the correlation coefficient for determining the state of this mode in the previous sampling period.

[0112] Example 2, Figure 2 The present invention provides a human state monitoring method based on low-power data feature extraction, comprising the following modules:

[0113] Low-power data acquisition module: used to acquire motion signals and physiological signals of the target human body to form an initial low-power data sequence;

[0114] Nonlinear Coupled Migration Prediction Module: Used to perform feature migration processing on the initial low-power data sequence to generate a migration coupled feature sequence;

[0115] Microstate enhancement feature generation module: used to enhance transient features sensitive to microstates based on transfer coupling feature sequences through a dynamic weighting strategy, resulting in an enhanced microstate feature sequence;

[0116] Power consumption feature selection module: used to dynamically select a subset of features from the enhanced microstate feature sequence based on the current power consumption level and sampling conditions, and generate a power consumption optimization feature sequence;

[0117] Human body state determination module: used to pass the power consumption optimization feature sequence through the human body state determination model to obtain the human body state determination result.

[0118] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0119] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0120] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0121] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0122] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0123] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A human condition monitoring method based on low-power data feature extraction, characterized in that, Includes the following steps: Collect motion and physiological signals of the target human body to form an initial low-power data sequence; The initial low-power data sequence is subjected to feature transfer processing to generate a transfer-coupled feature sequence. Based on the transfer coupling feature sequence, a dynamic weighting strategy is used to enhance transient features sensitive to microstates, resulting in an enhanced microstate feature sequence. Based on the current power consumption level and sampling conditions, a subset of features is dynamically selected from the enhanced microstate feature sequence to generate a power-optimized feature sequence. The power consumption optimization feature sequence is passed through the human state determination model to obtain the human state determination result.

2. The human body state monitoring method based on low-power data feature extraction according to claim 1, characterized in that, The process of collecting motion and physiological signals from the target human body to form an initial low-power data sequence is as follows: Based on the energy consumption and information mapping relationship of multi-mode sensing nodes, the energy consumption threshold range of the target node is determined. Based on the energy consumption threshold range, motion signals and physiological signals are asynchronously sampled to generate a cross-modal time mapping matrix. The transient response similarity of the sampled segments is calculated to select preliminary sampling blocks. Energy adaptive compression coding is performed on the initial sampling block, and sampling points with information entropy contribution below a preset threshold are set to zero to generate compressed low-power joint data fragments. Multiple low-power joint data segments are spliced ​​together in chronological order and then interpolated and smoothed to form an initial low-power data sequence.

3. The human body state monitoring method based on low-power data feature extraction according to claim 2, characterized in that, The asynchronous sampling of motion signals and physiological signals based on the energy consumption threshold range is specifically as follows: Motion signals are continuously acquired using a first sampling mode, wherein the motion signals include at least one of acceleration and angular velocity components; Physiological signals are acquired using a second sampling mode, which is triggered by a rate of change event of the motion signal. Establish a cross-modal time mapping matrix for calibrating the temporal relationship between motion signals and physiological signals.

4. The human body state monitoring method based on low-power data feature extraction according to claim 3, characterized in that, The step of performing feature transfer processing on the initial low-power data sequence to generate a transfer-coupled feature sequence is as follows: Identify the sampling gap regions of each modal signal in the initial low-power data sequence and determine their timing boundaries; Construct a cross-modal cross-mapping model with motion signals as the driving variable and physiological signals as the target variable, and calculate the transfer weights; Based on the transfer weights, transfer inference and calibration of the target modal signal are performed to form a transfer coupling feature sequence.

5. The human body state monitoring method based on low-power data feature extraction according to claim 4, characterized in that, The method for identifying the sampling gap region of each modal signal in the initial low-power data sequence includes: Based on the disappearance of signal response and the low power consumption state of the device, the sampling gap region caused by low power sampling constraints is jointly identified; By referring to the multimodal co-sampling points before and after the sampling gap region, and combining the preset time offset coefficient and historical propagation weight, the temporal boundary of the sampling gap region is dynamically inferred.

6. The human body state monitoring method based on low-power data feature extraction according to claim 5, characterized in that, The enhanced micro-state feature sequence is obtained by using a dynamic weighting strategy to enhance transient features sensitive to micro-states based on the transfer coupling feature sequence. Specifically: The migration coupling feature sequence is normalized to form a cross-modal transient response vector, and the weight vector of each mode is obtained based on historical sample analysis. Based on real-time energy consumption constraints and signal fluctuations, the weight vector is dynamically adjusted to obtain the real-time weighting matrix; The migration coupling feature sequence is weighted and fused with the real-time weighted matrix to output an enhanced micro-state feature sequence.

7. The human body state monitoring method based on low-power data feature extraction according to claim 6, characterized in that, The weight vector of the mode is obtained through the following steps: Calculate the rate of change of the temporal gradient of each modal signal within the sliding time window; The modal sensitivity factor is determined based on the mutual information value between the gradient rate of change and the microstate label. The sensitivity factor is normalized to generate a modal weight vector.

8. The human body state monitoring method based on low-power data feature extraction according to claim 7, characterized in that, The step of dynamically selecting a feature subset from the enhanced micro-state feature sequence based on the current power consumption level and sampling conditions to generate a power-optimized feature sequence is as follows: Establish a mapping relationship between features and power consumption; Obtain the current power consumption status of the device to determine the allocable power consumption budget, and filter the set of candidate features that meet the budget constraints; Calculate the temporal correlation between each feature and the target microstate, and perform a comprehensive ranking based on multiple factors; Based on the ranking results, a subset of features is selected and fused to generate a power consumption optimization feature sequence.

9. The human body state monitoring method based on low-power data feature extraction according to claim 8, characterized in that, The process of passing the power consumption optimization feature sequence through the human state determination model to obtain the human state determination result is as follows: The instantaneous response weights of each feature component of the power consumption optimization feature sequence are calculated using a human state determination model, and a dynamic representation vector is generated based on the weights. The confidence level of the microstate is calculated based on the dynamic representation vector, and a re-determination process is performed on the low confidence results to output a stable human state determination result. The re-determination process includes verifying the current state using historical determination results or supplementary features.

10. A system for human state monitoring using the low-power data feature extraction method as described in any one of claims 1-9, characterized in that, Includes the following modules: Low-power data acquisition module: used to acquire motion signals and physiological signals of the target human body to form an initial low-power data sequence; Nonlinear Coupled Migration Prediction Module: Used to perform feature migration processing on the initial low-power data sequence to generate a migration coupled feature sequence; Microstate enhancement feature generation module: used to enhance transient features sensitive to microstates based on transfer coupling feature sequences through a dynamic weighting strategy, resulting in an enhanced microstate feature sequence; Power consumption feature selection module: used to dynamically select a subset of features from the enhanced microstate feature sequence based on the current power consumption level and sampling conditions, and generate a power consumption optimization feature sequence; Human body state determination module: used to pass the power consumption optimization feature sequence through the human body state determination model to obtain the human body state determination result.

Images